This PDF file contains the front matter associated with SPIE Proceedings Volume 12168, including the Title Page, Copyright information, and Table of Contents.

The 3D modeling technology can show the bedroom space environment more completely, giving a more realistic sense of participation and experience. The use of three-dimensional modeling technology design can intuitively display the real environment of the bedroom space functional area in the future, so as to facilitate the timely adjustment of the existing problems, such an integrated approach to rationalize the time while enhancing the efficiency of work. The functional area of the bedroom space is no longer single in the traditional bedroom space on the layout of the dynamic line to optimize the design of the ideal state of the dynamic line. The combination of computer 3D modeling and V-RAY rendering technology makes the effect vivid and evocative, and better reveals the optimized dynamic state.

With the help of computer modeling technology, furniture can be expressed through virtual reality technology in the design stage, which greatly reduces the design cycle and design cost in the furniture design industry and improves the design efficiency. On this basis, the application of computer modeling technology in home design is discussed, and the application of furniture design is studied from the perspectives of ergonomics, formal beauty law and function.

One of the important measures for ship safety is ship detection. Due to the limited visibility at night, it is crucial to detect the ship's infrared night-vision image. We updated the YOLOX to enhance the precision and real-time of ship detection at night, primarily by replacing the CspLayer with a Ghostbottleneck that could adjust to a simple infrared image of the sea environment. Based on the public ship infrared image datasets, our configuration is trained and evaluated. Experiments reveal that, compared to the present algorithm, our method can expedite convergence rate, enhance the F1-score of ship infrared image detection by 0.02, and increase the mPA value by 2.93%.

Small scale damaged image inpainting is divided into iterative inpainting and interpolation-based image inpainting algorithms. In previous studies, it is found that there are some phenomena such as sawtooth effect, edge blur and so on. The sawtooth effect exists because the weight calculation of neighborhood pixels only considers the spatial information, ignores the image feature information. In addition, there are many factors leading to edge blur including the error accumulation of inpainted pixels, the low weight of gradient direction, and the inpainting priority of pixel. In this paper, we propose an image inpainting method based on Fast Marching Method (FMM). Firstly we use gradient direction and isophote direction to maintain the edge consistency, so as to improve the algorithm effect. Secondly we explore the confidence matrix to reduce the error accumulation. Finally we use the low rank of image to preprocess the image region to reduce the algorithm time complexity. The results of the experiment show that our improved method get well performance in inpainting procedure than the-state-of-the-art method.

This paper contains four major concepts about the quadcopter. Firstly, I drive dynamics analysis of the quadcopter. Mainly, apply force analysis with Newton and Euler’s method on the quadcopter. Then, applying first order ODE and second order ODE conduct formulas to obtain velocity and location expression. This part also helps to construct the computer simulation. Secondly, this paper introduces a commonly used quadcopter control technique-PD control. PD control stands for proportional-derivate control. It’s a control method that measured the desired position and actual position then outputs control commands. Third, the paper illustrates how machine learning algorithms can be applied to the control of the quadcopter. The machine learning algorithm used in this paper is the deep deterministic policy gradient (DDPG). Finally, I designed three tasks for experimental purpose. These tasks include flying directly upward from the ground, flying in x-direction, and flying in all three directions at the same time. Then, evaluating how does each control method works through these experiments.

Human action recognition in videos has attracted much attention due to its wide application prospects in research field, such as human-computer interaction, intelligent detection and video search. And deep learning technology has achieved very good results in many fields of the computer vision, especially convolutional networks have achieved the best results in image recognition. However, most convolutional networks are not suitable for dealing with temporal problems. In response to the problems, this paper proposes a new computational framework for the problem of human action recognition from videos, which is presented in this paper by using modified deep residual network. First, the deep residual network is used to obtain features through the fully connected layer, then the recurrent neural network with the residual structure is introduced to improve the efficiency of classification. According to the similar characteristics of adjacent frames in content, the residual structure is introduced into the RNN to increase the depth of the model and obtain better classification effect. Moreover, in order to solve the over-fitting problem, pruning with preference is carried out. According to the proportion of parameters of each node, some parameters are cut off in the training process. Data experiments on some most popular benchmarks as Olympic sports and HMDB51 demonstrate that the presented method could achieve comparable good results to the state of the art.

In the nature and wildlife protection field, biologists want to learn about endangered animals' behavior, environment, and activities. The hope is to get more information about endangered animals, such as their diets, predators, and surroundings. With the help of digital devices, biologists use drones to capture information about the changing environment of endangered animals. Therefore, high-resolution images with precise details are essential in the research process. Recently, there have been many deep learning models for improving image resolution in the computer science field. For example, super-resolution convolutional neural networks (SRCNN), efficient sub-pixel convolutional neural network (ESPCN), deeply-recursive convolutional network (DRCN), and super-resolution generative adversarial network (SRGAN). SRGAN is a deep learning network model based on generative adversarial networks. It has a good effect on improving image resolution. Based on the SRGAN model, the paper refined the loss function and structure of the SRGAN networks, which optimized the details of the generated high-resolution images, and provided more accurate high-resolution image results. The refined SRGAN aims to give better results in animal image analysis. We selected DIV2K dataset as the training and testing data and compared the refined SRGAN method with the original SRGAN method. The quantitative evaluation used the peak signal-to-noise ratio (PNSR) and structural similarity index measurement (SSIM). Compared with the SRGAN model with one convolution layer, the model improves the average PSNR value of 0.93dB, and the SSIM value reaches 0.8723. With the help of the improved SRGAN network, biologists can obtain better super-resolution images under the premise that other conditions remain unchanged to protect wild animals.

Different from convolutional neural network, transformer is able to model the long-distance relationship between the image pixels, thus it is now widely used in computer vision and remote sensing community. This paper comprehensively reviews the development of transformer models in automatic image interpretation tasks, especially the applications in image classification, object detection and semantic segmentation. Specifically, the popular transformer models are thoroughly analyzed and compared to acquire their advantages and limitations. Finally, current challenges and future works are concluded.

With the development of neural networks, the effect of single-image super-resolution is getting better. However, in many models that have been proposed, the processing of the upsampling module for image magnification is relatively simple. In this paper, we propose an iterative up and downsampling residual network. We selected a baseline model, and combined the upsampling sub-module and the downsampling sub-module into the final image upsampling module in an iterative form. This module generates detailed differences of the image by connecting convolution and deconvolution, and then extracts them through the residual method to correct the texture of the upsampled image. The iterative module enhances the effect of correction. Experiments on public datasets prove the effectiveness of the model.

Predicting retail sales is a hot research topic that can help firms achieve on-demand procurement. That can reduce the extra costs caused by an inventory shortage or surpluses. Traditional methods usually regard the sales of a certain product as the independent time series and then solve the task by a curve fitting model. However, the sales volume is not a single value, but the one aggregated from many types of products. There are correlations between different types of product sales. The increase in one type of product sales may lead to changes in sales in another type. Without capturing the correlation between different product sales, it is difficult to obtain satisfactory performance. To solve this problem, we propose a new multi-task method based on multivariate time series learning. Considering there are multiple product types and each type has different trend characteristics, we first cluster the time series of all product types and then model each as a multivariate prediction task. We then design a new dilated convolution network to fusion the features of related products and the trend characteristic of each task. Moreover, we develop a trend structural entropy network to grasp the fluctuation features of the task. A new self-enhancement mechanism is proposed to finely capture the correlations among tasks. Through multi-task learning, the model can effectively improve prediction accuracy by using the complementary information of closely related time series. Experimental results on two real datasets show the effectiveness of our approach.

Machine learning can be applied to obtain detailed information about wild giant pandas in a specific area, such as number, age, and gender; and using Maya to create a similar environment and adjust the pressure and sunshine. According to the previous work and experience, the aforementioned statistical work on the number of wild giant pandas in China is quite a time-consuming task up to three to four years. However, it is too late to formulate protection plans and strategies from the observation and exploration of pandas. The current development of machine learning will allow scientists and researchers to attain valuable and reliable information automatically analyzed, and even there is no need for researchers living in the forest. Here, we mainly demonstrate the application of the footprint identification technique which the computer divides the picture into two groups with footprints or not and then identifies which footprints belong to pandas from all animals. Furthermore, it will be capable of classifying and clustering their number, sex and age. It is based on the algorithm of k-nearest neighbor and clustering collected panda images. Moreover, it is also can be applied to another animal with a free-ranging living habitat seeking conservation without the invasion of a human being.

Accurate extraction of the height of tanks based on remote sensing images is very important in engineering applications. An efficient scheme to extract the height of oil tanks based on a single high-resolution satellite images with same-name-arc-distance (SNAD) is proposed in this paper. Firstly, image enhancement and geometric correction are carried out on remote sensing images with oil tanks, which makes the image not only have spatial geographic information, but also sharpens the boundaries of the shadows of oil tanks. Secondly, the technology of automatic detection and interpretation by experts for the boundaries of oil tanks’ shadows was used to identify and locate the same-name-arc-edge (SNAE). Thirdly, SNAD was proposed and the value of SNAD was calculated using the coordinate model of same-name-arc-point (SNAP). Finally, the heights of oil tanks can be extracted by using a new cylinder height measuring model. In order to verify the validity and feasibility of the proposed scheme, four sub-meter satellite images at different times were used to measure the heights of the same oil tank. Experiments show that SNAD is not only easy to interact and operate, but also has high precision and low computation complexity. Therefore, it would be very useful in engineering applications.

In the design process of space engineering R & D , engineers need to use a lot of existing knowledge to assist design. A modeling method of space engineering knowledge network based on ontology is proposed relying on existing Aerospace Thesaurus. Through knowledge resource supply-demand matching technology and knowledge network map, a knowledge push method for space engineering is designed combined with specific tasks. It has been preliminarily applied in the Space Engineering Data Center Platform. The tests shows that this technology can effectively improve the standardization of aerospace terminology and the accuracy of knowledge service. It also promotes space engineering R&D knowledge’s effective organization and utilization.

The current single image rain removal method is not ideal in the task of rain and haze removal. In this paper, based on the physical imaging mechanism under heavy rain conditions, a feature enhanced composition model, FEC-Net, is constructed by combining physical model-driven and data-driven methods. The feature maps, which contain rain streaks, atmospheric light, and transmission map information were obtained by three convolution layers. These feature maps are enhanced by the feature enhancement module. Finally, a clear image is obtained according to the formula of the atmospheric scattering model. The selective enhanced processing of key output data greatly improves the processing effect of the network. Extensive experiments show that our FEC-Net achieves better accuracy and visual improvements against state-of-the-art methods.

Since human visual properties have ambiguity, the neighborhood information of pixels in an image also has ambiguity. Using fuzzy theory for image enhancement allows mathematics to learn the advantages of the human brain in recognition and judgment of ambiguous phenomena, and use precise mathematical methods to deal with things that cannot be accurately represented by mathematics. In the article, the basic concepts of fuzzy theory, several common fuzzy distributions are introduced, and image transform domain enhancement algorithm based on fuzzy theory are studied. It is shown that the fuzzy theory algorithm can improve the contrast of useful information of images and facilitate people to get useful information more quickly.

In social life, social circle has always been a concern of people, and it is also a hot research issue in computer and other disciplines. With more and more social network users, the amount of data on the network is growing rapidly, and the scale of graph data is becoming larger and larger, which leads to the problem of computing social networks more and more difficult. SparkGraphX is a very important component in the spark framework and has great advantages in distributed computing and iteration. Using the built-in graph calculation function in SparkGraphX, we can quickly find everyone's social circle and greatly improve the calculation efficiency.

Deep learning based generative models can generate pictures with desirable fidelity and quality. In this paper, we implemented CycleGAN that doesn’t rely on paired datasets in animation industry to transform natural landscape pictures into Japanese animation style pictures. As demonstrated by a set of comprehensive benchmarks, we assume CycleGAN may have the potential to upend the whole animation industry. Numerous results on our dataset show the effectiveness of the proposed method. Our method finally obtains 0.9877 of PSNR and 17.1522 of SSIM, and we also visualize the output results of our images. Our method can give a brief attempt of image style transfer, which may be widely applied to many other different areas.

With the rapid development of deep learning technology and medical technology, neural networks are widely used in the field of medical image segmentation. Among them, U-Net neural network has gradually become a research hotspot in the field of image segmentation because of its good segmentation performance. It provides doctors with a consistent method of quantifying lesions and is widely used in the field of medical image semantic segmentation. This article studies the U-Net network, learns theoretically from the U-Net network model and its basic principles, and then conducts experiments on three typical medical images of liver medical images, fundus blood vessel images, and lung nodule images to explain various types of medical images. The characteristics of the image and the difficulty of segmentation, and the performance of the U-Net network in the relevant image segmentation is verified. Finally, the problems existing in U-Net network are discussed, and the future development is prospected.

Because of the interference of noise, the accuracy of image recognition will decrease. Therefore, an automatic recognition method of multi-commodity image in Internet-of-Things system is proposed. The local pixel block matching algorithm is used to construct the PCA training sample set, and the principal component analysis is used to model. The model is used to filter the noise of commodity image, and the epipolar constraint is introduced into RANSAC algorithm to ensure the reliability of recognition results. Experimental results show that the proposed method can achieve more than 95% accuracy when the threshold of the noise interference strategy is 30%, which is obviously better than the two comparison methods.

The performance index of artillery weapon sight calibration directly affects the probability of first shot hit. In order to improve the accuracy of artillery image stabilization calibration, a highly robust artillery image stabilization calibration method was established. Through the establishment of a long straight line fitting model for edge detection, the calibration of the gun’s image stabilization bullseye, adaptive fine mask region of interest extraction and local intersection extraction, realize image stabilization and precise calibration detection. The e experiment shows that the calibration method designed in this paper reduces the average calibration error of 0.5292 compared with the existing algorithm, and can more accurately realize the calibration and detection of the gun sight.

In order to improve the visual effect of the image and solve the problems of long image correction time and low accuracy of image correction in traditional methods, a real-time correction method for image distortion in virtual reality scenes is proposed. The deep learning method is used to classify the virtual scene. After determining the type of virtual scene, the Hough transform method is used to extract the image distortion target contour, and the edge detection algorithm is used to make the position of the target edge reach the sub-pixel level. Based on the edge distortion detection, a distortion model that can describe the image distortion is constructed, and the image distortion is corrected. The experimental results show that the correction time of this method is short, the correction accuracy is high, and the visual effect of the corrected image is good, which fully verifies the effectiveness of this method.

This paper introduces a novel method to refine the features in a deep learning (DL) model which is built for medical image classification. By framing it as a fine-grained visual categorization task, we propose an attention mechanism based on the inner product between feature column vector and the learnable attention vector. The proposed method is conducted on two open-source benchmark datasets. The results of the experiment suggest that the model be quite stable and accurate under the indicators of accuracy, precision, recall, and F1-score. It is found that inserting the inner product attention not only enhances the overall classification accuracy and efficiency, but also improves the recognition ability of learned features. Indeed, employing the proposed attention mechanism can increase the feature extraction capability of the DL models.

The accuracy of traditional image recognition methods is low, and the image recognition effect decreases. Therefore, an accurate target recognition method of transmission line foreign object image based on deep learning is designed. The foreign object image data of transmission line are collected, and the network structure of foreign object image of transmission line is extracted. Based on deep learning, the accurate index of foreign object image recognition of transmission line is optimized to realize the accurate image recognition. By means of comparative experiment, it is verified that the image recognition effect of the new method is better and has great popularization value.

The existing research on software testing visualization is mostly at the code and developer level, and the visualization content is mostly at the code line and document level, and the target user group is very limited. In addition, the thirdparty software evaluation agency involves a large number of paper documents in the testing and review, which is inconvenient to manage, difficult to consult and analyze, and work efficiency is low. In order to solve these problems, we propose a visual analysis framework for software functional testing process. Starting from the functional testing process and the key information of each process, a set of visual analysis process is constructed, the data extraction method of the test document and the visual analysis method of the test process are studied, and the visual analysis prototype system is constructed to support the software testing process. And the key information of each process for quick analysis and further exploration. Finally, we demonstrated the application of the system in actual test projects through case analysis.

In this paper, a distributed machine learning algorithm is proposed to optimize the placement of in-network computing resources in network computing by redesigning aggregator hash mapping. Based on this placement method, the three models were tested, and the training task throughput increased by 6.81%, 0.39% and 3.13%, respectively, while the task completion time decreased by 6.82%, 0.28% and 3.31%, respectively. At the same time, based on the placement method, the network computing training system can support more flexible network computing resource allocation mode.

Aiming at the problem of uneven gray distribution and low contrast in medical images, the paper proposes a medical image enhancement method based on histogram equalization. The image is initially improved through wavelet transform combined with threshold processing. The detail features of the improved image are extracted by Prewitt operator and fused with the original image, and then the histogram is used to equalize and enhance the image. Experimental results show that the proposed method can better enhance the image feature information and is suitable for a variety of low-quality medical images.

In contemporary society, convolutional neural networks (CNNs) are used mainly for image processing, classification, segmentation, including detecting skin cancer. However, many current CNN models used to analyze skin cancer heavily rely on enough information apart from photos of skin lesions. This need hinders the users from diagnosing themselves. To tackle this issue, the paper proposed a new but simple CNN model. It can be utilized to detect and differentiate benign and malignant skin cancer on the patients’ end using image classification. Deep convolutional neural networks, such as CNN, demonstrate the potential to be properly trained and making predictions based on the given training set. In this paper, we show classification of skin cancer using a single CNN model, trained end-to-end from real skin lesion images directly. It only uses pixels and disease labels as inputs. We used the malignant vs benign skin cancer dataset generated by Claudio Fancon. Images are provided by the public ISIC (International Skin Image Collaboration) 2018 Skin Lesion Dataset. Furthermore, we resize the original images to a desirable resolution that is ready to be used for training. Then, we build a multiple-layer CNN model and train it on all the data. Finally, we fit the model to a local user interface for future use. The proposed model is found to be successful and promising, achieving testing results with 82% accuracy on the test set. This could significantly reduce human mistakes in the skin cancer diagnosis process. It also obtains an average loss of 0.45, which is a relatively low figure on the test set. The proposed system, therefore, is relatively reliable and robust when detecting potential skin cancer.

Omission of entities in the extraction process is caused by the model that performs joint extraction of entity relations by estimating the initial and last positions of entities. The probability threshold for predicting entity positions is dynamically altered in this work by modifying the probability threshold for judging entity positions. This reduces the loss of subject entities in the prediction process based on assessing entity positions using the proximity principle. Furthermore, different weights are provided to distinct entity relations to optimize the model's effect on entity relation extraction with few samples and raise extraction accuracy. Experiments show that, when compared to the existing algorithm, the algorithm can reduce the probability of the model missing subject entities during the extraction process, and the model's accuracy improves by nearly 0.2% to 0.5% in a small number of samples extracted using random sampling. The accuracy of model extraction can be improved by altering the weight assignment function, and the experimental results have improved by approximately 1%.

Under the background of full-scale construction of smart grid, demand side resources will play an important role. Through game theory, the demand side resources will be mobilized to connect with the power market, and Nash equilibrium principle of game theory will be applied to establish an equilibrium point between power plant, power grid and user demand, so as to maximize the benefits and the maximum consumption of new energy. According to the application of game theory in demand response, this paper aims to solve the problem of dweller aggregation load, establish a mathematical model.

Nowadays, data mining is not only used for scientific research, but also has various applications in real life. Support vector machine (SVM) is a kind of data mining method and a generalized linear classifier for binary classification of data. Its basic principle is to map vectors to a higher dimensional space, find the maximum spacing of the separated hyperplane in this space, and this separated hyperplane is unique. In this paper, the characteristics of credit card holders and the transaction data of six months are converted into data formats required by data mining, and the prediction model of support vector machine is established. According to the predicted results, credit card default risk is analyzed to further evaluate the model used in this paper.

This article focuses on the treatment of a serious congenital heart disease: tetralogy of Fallot. According to the existing treatment of tetralogy of Fallot, patients undergo multiple thoracotomy operations to implant catheters to adapt to different physical conditions, and young patients have to bear the risks and pains. Therefore, this study intends to improve the feasibility of existing treatments: according to the body data of each patient, the implanted catheter is designed personalized into a pleated structure, so that it can deform according to the increase in blood pressure and meet the blood pressure of different ages and reduce patients’ surgery. In the test method, the computer software SolidWorks is used for modeling, and 3D models of different design schemes are produced; and then the models are imported into the computer software Abaqus. The finite element method is used to set the material and blood pressure parameters, analyze the deformation of the catheter under various blood pressure conditions. The conclusion shows that the growth rate of catheter deformation decreases with increasing force, and gradually tends to a steady state; the number of folds and the angle of folds have a significant impact on the degree of catheter deformation.

To analyze Carboniferous oil and gas resources in Xiquan 103 well area, Petrel modeling software is used to integrate the previous geological research results, According to the geological structure characteristics of volcanic rocks, the modeling ideas are reasonably formulated, three-dimensional geological model of dual-medium reservoir of volcanic rock is established by establishing key steps such as structural model, reservoir facies model, matrix reservoir parameter model and fracture model, so as to realize three-dimensional quantitative prediction of matrix and fracture reservoir parameters. On the basis of model verification, reasonable coarsening grid provides reliable three-dimensional geological model for numerical simulation. The accuracy of the model established in this paper is 25 × 25 × 0. 5m, and the total number of grids is 5.81 million. The model verification has high coincidence, and the reserves are calculated by units, and the fitting error of reserves is less than 5%. In this paper, Carboniferous oil and gas resources in Xiquan 103 well area is studied and implemented, which provides reference for subsequent oil and gas field development.

Eutrophication is the most common water environment problem of receiving water body. As a new literature analysis method in recent years, visual analysis of literature can be used to obtain some profound and intuitive understanding of the research content and development trend of eutrophication models. In this paper, literatures of eutrophication models in the past decade are analyzed from several different angles by CiteSpace, a visual analysis software. Results show that the international research on eutrophication models is gradually increasing. Research hotspots in this field mainly focus on lake eutrophication, Lake Erie, marine hypoxia, Baltic Sea, life cycle assessment, ecological degradation, Black Sea and so on. Research fronts in this field mainly focus on cyanobacteria and eutrophic lakes. According to the selected time interval, it can be divided into three time periods. The preliminary research focuses on eutrophication at the estuary and offshore; the medium-term research focuses on cyanobacteria lakes, and the later research focuses on river eutrophication. China hold the largest number of eutrophication research, which are mainly originated from the Chinese Academy of Sciences. In addition, China has close cooperative relations with many countries in the field of eutrophication models study which cross with many other research fields.

Newly built EV charging stations are usually equipped with battery storage system and PV panels, which can maximize the usage of cleanable power energy, and provide great capabilities for meeting various control objectives of the power grid. Operated properly, the controllable load of such charging stations can be used to provide ancillary services in an electricity market. This paper presents a novel predictive optimization method that adopts deep neural network (DNN) for forecasting EV loads and electricity market prices and optimally operating storage systems. The DNN-based forecasting methods consist of key function modules of data collection/cleaning, feature engineering, model training with parameter tuning, and adaptive updates for ensuring long term effectiveness. The economic operation of an EV charging station is formulated as a nonlinear multi-objective optimization problem with multiple constraints. Detailed mathematical optimization models considering time-of-use electricity and peak shaving market are provided. Comprehensive case studies are conducted on an actual EV charging station using real operation data that verifies the effectiveness of the proposed models and methods.

This paper proposes a method to evaluate the time to collision (TTC) by the flow traveling through the road work zone based on the traffic conflict technology. The reliable big data are extracted from observed videos and the generalized exponential function is designed to explore the probability of the potential collision. The probability of rear-end collision when vehicles passing by the designated work zone is discussed in detail, in specific, in the scenario of an urban four-lane road work zone. Finally, four macro factors are exhibited, which have significant impact on the TTC including observed speed, flow, distance and following vehicle type. Conclusions are drawn that the vehicle speed is faster, the traffic volume is greater; meanwhile, the closer to the work zone area, the rear-end collision probability is higher.

With the development of new generation information technology such as big data analysis, artificial intelligence and image analysis, intelligent detection and identification of distribution network line defects has become possible. At present, the identification of distribution network line defects mainly depends on the human eye observing the visible image, but the visible image is easy to be affected by uncertain factors such as jitter or defocus, resulting in identification errors. In order to make up for the shortcomings of the existing recognition technology, this paper combines the infrared image with the visible image, and designs an image enhancement method to image the defects of the distribution network line based on the "infrared + visible" dual vision system. On this basis, the infrared image features are extracted, the defect database is constructed, and the distribution network line defect recognition algorithm based on image enhancement is designed to detect and recognize the equipment defects through multi network fusion.

Because of the underwater lighting and some impurities in the water, the images taken under the water tend to become very blurry and the color is very dim, this paper uses the matlab R (red), G (green), B (blue) tricolor layer, its grayscale and grayscale value as histogram and with some algorithms to do the grayscale histogram of the equalization of the original blurred underwater image into a bright color image, and with the removal of noise filter to soften the image. to enhance the underwater image. And this paper through simulation experiments to prove that the image itself can be preserved and get good image enhancement results.

Different from the existing empirical landscape measurement methods, this article explores an effective method to quantify the landscape through the use of computer vision perception. First, based on image semantic segmentation technology, a large number of excellent case images are put into image semantic segmentation and annotation tools for image segmentation. Secondly, the image semantic segmentation is used to quantitatively analyze the landscape space elements of the old urban site, and obtain the data value of each landscape element in the excellent case, which provides a reference for the design of the element ratio data basis. Finally, it is applied to the street space in the old city of Macau to perform a quantitative analysis of the computer vision perception of the streets in the old city. Get quantitative suggestions for landscape improvement, and provide new methods for quantitative research and transformation of old urban areas.

Mass customization is the direction of China's solid wood furniture companies in the future. Group technology can solve the problems of multiple categories and small batches brought about by mass customization. The rational part division is the prerequisite for the application of group technology. This paper is based on fuzzy cluster analysis and BP neural network technology to study the clustering of solid wood furniture parts into groups. The case analysis conducted experiments with 164 parts of solid wood furniture from a company, and the results showed that 2-8 part families are better GT solutions. After 8 iterations, the error of the BP neural network is reduced to less than 1*10-6, the parts participating in the test are accurately grouped, and the neural network meets the requirements of actual use. The case verifies the method proposed in this article, and the results show that this method is a method of grouping parts suitable for the production of solid wood furniture, which can help solid wood furniture companies cope with mass customization production.

Current relational triple extraction approaches rely on a large amount of labeled data which is hard for many real-world applications. How to extract relational triples has become a concern in few-shot settings. At present, the researches in few-shot settings are rare on the joint extraction tasks. The main problem is that the accuracy of subject and object recognition is relatively low. To improve the accuracy of subject and object recognition, we propose an end-to-end few-shot relational triple extraction model with nearest neighbor matching. Specifically, we recognize subject and object in the sentence according to the semantic similarity of words, and propose the new method to discriminate the subject and object. Also, we add the loss of the subject and object type as a penalty item for model training. Experiments results demonstrate that our method greatly improves the performance of subject and object recognition on the public English dataset FewRel, and achieves the state-of-the-art on the few-shot relational triple extraction task.

With the blooming of online video applications, live commenting is an emerging feature of online video sites. The live video comments generation (LVCG) task aims to generate live comments for videos while considering both the video and the surrounding comments made by other viewers. In this work, we aim to improve the relevance between live comments and videos by modeling the cross-modal interactions among different modalities. To overcome the problem of insufficient multimodal interactions for live video comments generation, we built two basic attention blocks: the self attention (SA) block that can model the dense intramodal interactions; and the x-guided attention (XGA) block to model the dense intermodal interactions. After that, by modular compositions of the SA and XGA blocks, we propose different multimodal transformer architectures to handle the multimodal features. Finally, experiments show that our proposed multimodal guided attention models significantly outperform previous methods in most of the metrics.

Taking 40 towns in Yancheng City as sample cases, this paper discusses the multiple concurrent factors and complex causal mechanisms that affect the high-quality development of towns by using fsQCA method. Based on the five antecedents of industrial development, personal income, ecological governance, party building and life experience, the driving path of high-quality development of towns is obtained. The results show that there are two types of "Party Building*Personal Income" and "Party Building*Ecological Governance", and there are four configuration paths in total. Among them, party building, personal income and ecological governance are the core conditions, which have a significant driving effect on the high-quality development of towns in the region. This research conclusion provides decision support for other towns to implement high-quality operation.

Recognition and classification of emotion is an important research direction in emotion computing. At present, the research of the emotional recognition focus mainly on vision and voice, and the accuracy of the recognition and the classification is very low, which is not enough for commercial application. Therefore, in recent years, the electroencephalogram (EEG) emotion recognition research has become an important direction in the emotion computing. The EEG emotion is controlled by the central nervous system, it is not easy to be controlled by subjective consciousness, and can reflect the objective and the real emotion. Therefore, this paper proposes an EEG emotion classification model based on one-dimensional convolution neural network (conv1d) and a gated recurrent unit (GRU). In the experiments, this paper collected the EEG emotion signals with the open source brain-computer interfacing (OpenBCI). After collecting seven kinds of emotion data, the collected EEG signals are sorted into a training set and a test set. After data training, the accuracy of the emotion classification in the model can reach more than 90%. This model has high accuracy and simple data processing.

Achieving image-to-image style transfer using various types of generative adversarial networks is a popular question in computer vision, with numerous works being proposed. However, the current typical research on the use of neural networks to achieve image-to-image style transfer is to improve the output performance through the improvement of the model, and the intermediate process of style transfer has not been paid much attention. In this paper, based on the implementation of CycleGAN style translation, the gradients of non-leaf nodes are calculated and extracted by using the autograd function in PyTorch; and finally, the aging images of faces at different ages are dynamically generated by using the middle layer gradient. The experimental results, based on the AGFW-v2 dataset, tell us that this is capable for the initial research purpose; and the presentation of the process of style transfer intermediate enriches the stage changes of age. The processing of intermediate layer features in this study can also be used in areas such as photo enhancement, image dehazing and face swapping to achieve controllability of model processing results.

Face recognition technology has attracted much attention in the past few years because of its potential value of both research and applications in the fields of pattern recognition, image analysis, computer vision and other fields. Meanwhile, as an important branch of the deep learning family, convolutional neural network (CNN) methods has yield remarkable achievements in many large-scale recognition tasks in the field of computer vision, especially face recognition. This is because, compared to classic neural networks, CNN has more hidden layers, a more complicated network topology, and more robust feature learning and feature expression capabilities. This paper gives an overview of the development on face recognition using CNN approaches, including categories of current face recognition technologies, description of the CNN method, and applications and research progress on CNN for face recognition.

Artificial intelligence not only brings great convenience to the life and study of ordinary people, but also brings hope to special groups such as autism patients. Computers can process digital signals and images to present idealized intervention facilities in the treatment and rehabilitation assistance of special populations. At present, robot has been applied in the field of early diagnosis and monitoring of children with autism spectrum disorder and guiding situational game interaction in foreign countries, and its effects are very significant, providing choice and reference for the future application of artificial intelligence to special groups.

To save costs of manual reviewing, a tool was designed for automatically checking academic integrity and content consistency of abstracts through analyzing the academic problems of abstracts from multi-dimensional testing indicators. First of all, the recognition of abstract knowledge elements can be achieved with the help of the Naive Bayes algorithm and the posterior probability correction method. On this basis, the consistency between the abstract and the text can be checked in combination with semantic matching and knowledge element matching. As can be seen from the experimental results, the F value of the model can reach 0.85 in the academic integrity checking of abstracts. At the same time, the checking granularity is refined to effectively distinguish the abstracts of varying quality.

Based on the development status of augmented reality technology, combined with the needs of hospital construction in the information age, and in view of the inconvenience of navigation indexing in large general hospitals, the basic method of system construction is found by analyzing the characteristics of augmented reality technology and system architecture. Simulate and analyze the behavior of related users, explore the pain points and difficulties of user needs, combine indoor positioning technology, and design action routes for users on the basis of related navigation algorithms, realize virtual information prompts on human-computer interaction interfaces, and provide medical treatment for hospitals. Provide more convenient navigation and guidance services with visitors.

DeepFaceLab, which is also known as AI face swap, is a new task with practical value in computer vision field that has become popular in recent years. As it can replace any face in an existing video with another face whichever you want without revealing flaws, swapping face technology is now becoming more and more welcomed by the entertainment, film and art industries, producing high commercial value. A number of works based on convolutional neural network or generative adversary network have been proposed to extract facial features and accomplish face swap. However, their network architecture can't complete the small face swapping very well and the effect of the generated video is also not ideal. In this paper, we train on the SAEHD model based on a single shot scale-invariant face detector. We use the scale-equitable-face detection framework to ensure that we can extract enough features on different scales for face swap. In addition, according to the effective receptive field, the anchors of different scales are defined on different feature maps, and we adopt the proportional sampling method to ensure that the sampling density of the anchors on different feature maps is consistent. By replacing the face by frame, we have achieved very good results on the DeepFaceLab task, achieving relatively small source and destination loss, respectively, with the real-time speed.

Convolutional neural network models in the field of computer vision are also emerging in endlessly. From LeNet in 1998, to AlexNet, which detonated the deep learning boom in 2012, to VGG^[1] in 2014, and ResNet^[2] in 2015, the application of deep learning network models in image processing is getting better and better. Neural networks are getting larger and larger, their structures are becoming more and more complex, and the hardware resources required for prediction and training are gradually increasing. Often, deep learning neural network models can only be run on servers with high computing power. Deep convolutional neural networks have improved the performance of multiple computer vision tasks to a new level. The overall trend is to build deeper and more complex networks in order to achieve higher accuracy, but these networks may not necessarily satisfy mobile devices in terms of scale and speed requirements. Due to the limitations of hardware resources and computing power, it is difficult for mobile devices to run complex deep learning network models. Efforts are also being made in the field of deep learning to promote the miniaturization of neural networks. While ensuring the accuracy of the model, it is smaller and faster. From 2016 until now, the industry has proposed lightweight network models such as SqueezeNet^[3], ShuffleNet V1-V2^[4-5], MobileNet V1-V3^[6-8], and InceptionV1-V4^[9-12]. These models make it possible for mobile terminals and embedded devices to run neural network models. Therefore, this article compares the mainstream lightweight neural network models, analyzes and contrasts various networks, and looks forward to the future development trend of lightweight neural networks and several directions that can be studied.

"Thin as paper and sound as chime" is the reputation of the two important characteristics of optical and acoustics of Chinese porcelain. On most online display platforms, porcelain is only displayed in two dimensions, and the sound display is relatively single. This paper studies and proposes a simulation method for visual and auditory multidimensional interaction in a distributed environment. Based on the porcelain photo acoustics of virtual simulation, in the 3D simulation porcelain model, by adjusting the intensity coefficient of the light source, the light propagation model on the surface and inside of the porcelain is established to realize the thinness and light transmittance of the porcelain. This research combines the fundamental frequency, harmonic frequency technology and volume with the 3D porcelain model, and use its structural characteristics and the principle of chime to synthesize the percussion sound of porcelain in a virtual simulation environment. The research conclusions can be applied to fields such as art online platform display.

In the development process of computer software, artificial intelligence technology can establish network protection system with Agent technology; and can also build neural network detection system for intrusion detection with neural network technology to ensure that computer software can operate well. To further improve artificial intelligence technology, and strengthen the application of AI technology in computer software development, is also to further improve the speed and accuracy of network information security detection. In addition, combining with traditional intrusion technology, identify and intercept risks can also effectively reduce cyber security threat factors.

The continuous expansion of the scale of transmission lines has increased the workload of identifying potential dangers around them. Because the inspection cycle of long-distance lines is too long, the potential dangers are not found in time, which often leads to the damage of power equipment, resulting in large-scale power outages and huge economic losses. Therefore, the topic of transmission line peripheral hidden trouble identification based on image edge detection and monocular vision is proposed. Firstly, transmission data is optimized based on image edge detection, image edge detection data is optimized, and data pretreatment is carried out on transmission data. The identification model of hidden dangers around transmission lines is designed and the identification results are obtained by calculating the model. Calculating the shortest distance between the transmission line and the hidden danger, less than 3 km, through the alarm to remind the staff. Finally, the relative errors of distance calculation results are controlled about 2% by an example.

UAV patrol inspection is a new patrol inspection method to complete the maintenance of overhead transmission line by using unmanned aircraft for navigation and shooting and relying on the advantages of independent operation. Compared with the traditional aerial patrol, UAV patrol has the advantages of strong adaptability, high safety factor, low risk and low cost. However, there are image defects and other methods in the process of application. Therefore, in the face of this situation, it is necessary to design an image defect detection method for aviation insulators based on deep learning. Creating the insulator image acquisition and deep learning environment, determined the image segmentation detection threshold under deep learning, and located the insulator image defect detection fault. The Crop-MobileNet network detection model is established, and the CNN processing method is used to detect the insulator defects in aerial photography. The final test results show that under different deep learning convolution ranges, compared with the traditional scaling factor defect detection group, the deep learning image defect detection group has a higher recall rate. It shows that the defect detection effect of aerial images is relatively good, which has practical significance.

Intramedullary nail fixation is the gold standard for the treatment of lower extremity long bone fractures; however, intramedullary nails are often deformed after inserting into the medullary cavity. With the rapid development of computer graphics, the extraction and registration technology based on 3D point cloud is applied in the preoperative design. This paper presents a method for optimal position localization of intramedullary nail placement into the medullary cavity based on centerline registration. First, the femur and intramedullary nail were scanned using CT and reconstructed three-dimensionally. Second, centerlines of the medullary cavity and the intramedullary nail were extracted respectively. Curvature and best fitting diameter of each control point on the centerline were calculated. Third, the centerline of the medullary cavity isthmus was extracted. The extracted centerline was registered with the centerline of the intramedullary nail in each position. The curvature, diameter and distance measurement of each matching position was calculated. Finally, the best registration position was determined according to a weighted fitness at the individual slide positions. The proposed method was verified by experiments. The experimental results show that the obtained position is the best position for the protection of intramedullary osteogenesis. It is very meaningful to reduce the deformation of intramedullary nails.

At present, with the development of science and technology, artificial intelligence has increasingly become a common trend in life. For art and design, the popularization of artificial intelligence will also make a great breakthrough and change in the field of art and design. The combination of artificial intelligence and art design can provide more intelligent, scientific and creative design works for future design. Study the combination point between art design and artificial intelligence, analyze specific application examples, and explore the new expression and application of artificial intelligence in art design so as to open a new breakthrough for traditional art design.

Chinese herbal medicine plays an irreplaceable role in modern medicine at the moment. The recognition of the species of Chinese medicine is the precondition for drug development. The classification and recognition of Chinese herbal medicine relied on manual work, and the highest rate is about 70%. Most automatic recognition studies were based on two-dimensional imagines of Chinese herbal medicine for classification, while other studies used spectral information. These methods for kinds of Chinese herbal medicine was relatively single, and the band range was narrow, which ignored the three-dimensional structure characteristics. The accuracy, variety and efficiency of their recognition need to be improved. This paper first extract the spatial structure characteristic and ground object spectrum information of Chinese herbal medicine by 3D point cloud data and electromagnetic objects spectrum data. The classical classification model of deep neural network PointNet is used to classify the 3D point cloud data of Chinese herbal medicine, the object spectrum data is used to match and classify the spectral feature, and then verified by correlation coefficient method. This paper conduct a classification experiment by using the self-built data set containing seven kinds of Chinese herbal medicine. The results show that the overall accuracy of the test using PointNet for the automatic classification of seven kinds of Chinese herbal medicine at the same time can reach 86.3%, while the classification of the object spectrum verified by correlation coefficient method can reach 98%. Experiments have proved that the accuracy and the types of Chinese herbal medicine processed are superior to previous methods. This method can be used to identify Chinese herbal medicine and improve the production efficiency.

As an important part of object detection, rotated object detection plays a great role in many visual applications involving ship detection in remote sensing images, oriented scene text spotting, and face detection etc. Since the different definitions of the angle and the periodicity of the numerical angle values, the angular boundary discontinuity will happen during the model training which decreases the detection accuracy. To address that, we propose an arbitrary-oriented ship detection with RetinaNet embedded angle vector module (RetinaNet-AVM), which adds a new angle vector module (AVM), improves the boxes localization loss function and applies a new boxes angle loss function. At first, the numerical angle values will be encoded into angle vectors by the AVM before data training. After the training, the angle vectors will be decoded into numerical angle values to generate the bounding boxes of the final results by the AVM. On the HRSC2016 dataset, RetinaNet-AVM could detect the arbitrary-oriented, large aspect ratio and densely arranged ships with the rotated bounding boxes, and the precision is 90.76%, the recall is 89.66% and the average precision (AP) is 86.71%. Compared with the RetinaNet-H, our model has improved accuracy, recall and AP by 4.88, 1.47, and 3.9, respectively. And compared with R2 CNN, RRPN and RRD, our model is still competitive.

Video object segmentation (VOS) is a fundamental research area in the computer vision field in which the goal is to separate the target object(s) from the background in continuous frames of a video. It has huge application value and demand in human-computer interaction, video analysis, compression, re-creation, and numerous fields. Current VOS research in CV academia is mainly classified into four main categories: semi-supervised VOS, unsupervised VOS, weakly supervised VOS, and interactive VOS. In this paper, we give an overview of the latest methods in the first three categories. We summarized problems in each area and features of different methods aiming to solve them. Then we compare these methods to find out the performance in different test environments.

The structure of tire X-ray image is complex, and the defect information is highly related to the background information, which brings difficulties to automatic defect detection. To solve these problems, this paper proposed a tire defect detection algorithm based on multi-task learning and normal feature fusion. Firstly, a tire defect detection model was designed based on YOLOV4, and then a background classification task was added into the model using the idea of multitask learning. Finally, the result of model recognition is adjusted by fusion background feature detection algorithm. Experimental results show that the proposed algorithm can greatly improve mAP when detecting several kinds of defects with a large relationship with location, and has good application value.

Computer vision techniques are an important application for intelligent pest detection. However, it suffers from serious problems and challenges, especially distinguishing the targets of pests with high similarity, small size, and sample unbalance. In this paper, a domain-adaptive-calibrated-free-anchor detection network (DACFA-Det) is proposed, in which a balanced learning mechanism is added to the detection network, dealing with sample imbalance, feature similarity confusion, and center point inaccurate. Our method is evaluated on the re-established similar pest dataset (SPD). The final experimental results show that our method can obtain 44.0% mAP (Average Precision) on the SPD. The testing speed achieve 0.045s per image, meeting the real-time requirement, which proved the effectiveness and efficiency of our method for agricultural similar pest detection.

Detection on small objects is a tough task compared with large objects. It is even harder on smaller-size models because the accuracy might decline significantly. The main purpose of this paper is to explore and evaluate the performance of small object detection based on yolov5 series networks, including yolov5n, which is the latest released version and has a smaller size than yolov5s. Meanwhile, this paper will narrow down the general small object to small balls detection, including baseball, football, and tennis, as an example of general small objects in real life. The idea of this paper is basically to see if there is an easier way to detect missing little stuff in daily life, such as lost keys or chargers, to provide help to people who might have difficulty to see or find them. The method applied is to train the dataset containing the three balls sport involved and run them on yolov5n and yolov5s to evaluate their performance. The dataset collection is from Kaggle and labeled by this paper’s author, to simulate other small objects detection.

In a complex system, the strength and direction of feedback among different agents are difficult to quantified, especially due to the non-linearity. To quantitively estimate the coupling network of complex system, multiple statistical approaches have been widely adopted, including e.g., pairwise correlative analysis, Granger causality analysis. However, these analytic approaches often assume linear system and relative simple coupling pattern. In this work, I employed the concept of information entropy transfer and explore the non-linear interactions of near surface water and energy. This way could acutely find weak coupling effect of data and construct a chart of overview which can reflect the coupling impacts between variables directly. The goal of the analysis is to gain results about how energy circulate in the environment on the earth. Change in ways energetic transfer may imperil the survival of living matters on the path.I found that 1)From 1850 to 2005 the influences of existing variables other than the coupling relationship between precipitation(pr) and latent heat(hfls) are weak. 2)From the predictive model between 2006 and 2100,the effecting strength becomes relatively moderate and concordant.3)The time lag to surface temperature become affected by 3 variables(solar radiation,latent heat and sensitive heat) covering a whole year in the period between 2006 and 2100 in comparison with the other period with 2 variables(sensitive heat and latent heat) in the same month.

Coronavirus disease 2019 emerged in December 2019 and was later named Covid-19 by the World Health Organization (WHO). The disease is regarded to be the extremely contagious one with a high fatality rate worldwide. Therefore, the prediction and prevention of the disease are urgently needed. SEIR is a typical model of infectious disease propagation prediction. Hence, in this paper, we proposed a new model based on the SEIR model to predict the propagation of Covid-19 with the case study of the USA. Also, we use principal component analysis to evaluate the severeness of Covid-19 in each continent and propose an allocation plan of epidemic prevention supplies. Finally, the result of both models will be visualized by Python.

The occurrence of cold source blockage in nuclear power plant not only affects the safety of power production, but also brings challenges to nuclear safety. Cold source reliability has become another important bottleneck restricting the development of nuclear power. Based on the consideration of improving the cold source safety of nuclear power plants, this paper studies the intrusion events of several typical marine organisms in nuclear power bases at home and abroad, analyzes the causes of the events, establishes a defense system in depth for marine organisms, identification, detection, early warning and hierarchical response, and uses underwater acoustic high-resolution multibeam detection method to detect marine organisms such as jellyfish and hairy shrimp, so as to obtain their echo characteristics, analyze its distribution density in sensitive sea areas, and give effective early warning judgment. Among them, detection and early warning is the front-end disposal measure, which can provide emergency decision support for staff. Through the development and test of corresponding key equipment, the industrial safety risk is reduced, so as to improve the reliability of cold source.

UAV air combat game confrontation requires a high level of intelligence and autonomy, and needs the support of artificial intelligence technology to effectively improve the ability of autonomous air combat. This paper summarizes the reinforcement learning methods for UAV air combat confrontation. Firstly, the problem and application background of UAV air combat confrontation are introduced. Then, the research status of reinforcement learning methods at home and abroad is analyzed, including basic reinforcement learning, reinforcement learning based on Markov chain and deep reinforcement learning algorithm. On this basis, this paper focuses on the reinforcement learning method of UAV air combat confrontation from the two aspects of existing research results and algorithm application advantages. Finally, the simulation is implemented, and results show the effectiveness of reinforcement learning algorithm in air combat confrontation.

Aiming at the problem of jellyfish detection, a jellyfish detection method based on high-resolution multibeam acoustic imaging is proposed. The sea moon jellyfish detection experiments in real environment are carried out, and the individual echo characteristics of sea moon jellyfish are analyzed. The results show that the high-resolution multibeam acoustic method is effective to detect marine organisms at the outlet of nuclear power plant.

In this paper, a simulation model of Organic Rankine Cycle system (ORC) in an industrial park was established by using GREATLAB software. Analysis of Variance and sensitivity formula were used to analyze the influence of single factor parameters on the system. The results show that the system model built by GREATLAB has high accuracy, which can provide a reference scheme for the actual operation of the system. Through sensitivity analysis, the temperature of the heat source would have a great impact on the net output power of the system. The main factor affecting exergic efficiency was the temperature difference between the working fluid and the heat source.

At present, the wear of the electric collector shoe sliders of subway vehicles is mainly detected by naked eyes or simple rulers. There are problems such as low efficiency and low acquisition, and it is different to meet the needs of engineering applications. In this regard, this paper proposes a method for detecting the thickness of the electric collector shoe slider by the deep learning. First, establish the collector shoe slider dataset for the segmentation neural network method; second, use the attention mechanism and sub-pixel variable layer to optimize and train the original DeepLabV3 + neural network; finally, the optimized neural network and the traditional neural network method are used for comparative experts. The results show that the method proposed in this paper can significantly import the acquisition of the wear detection of the collector shoe slider.

For the interaction of new media art, it is a comprehensive art design, which involves a wide range. In the traditional interactive design of new media art, objects and models are often used as display carriers, and the commonly used auxiliary display means are the combination of words and images. This simple static display with material form as the medium can no longer meet the needs of current new media display design. In this case, it is very important to pay attention to the application of new technologies in new media display design. In this case, it is an important task to research and develop a new display technology to meet the needs of new media. Therefore, virtual reality technology came into being. This paper studies the interactivity of new media art based on virtual reality technology. To realize the mutual combination of technology and art, it is necessary to accurately grasp the cultural connotation of virtual art in the new media environment, integrate excellent virtual technology and art, and organically combine it with the contemporary digital trend; and finally create works that not only meet the time but also meet the aesthetic needs of the masses and the development of the times.

At present, scholars have studied the spillover effect of pro-environmental behavior, but its mechanism is not clear. This study records, identifies, analyzes and compares consumer behavior data through artificial intelligence model to predict the possibility of consumers' subsequent pro-environmental behavior. It is found that the spillover effect of pro-environmental behavior is also affected by cost adjustment variables, which provides practical guidance for the promotion of pro-environmental behavior.

Aiming at the language barriers of cross-border assistance, international communication and outbound travel of medical staff under the current and even long-standing pneumonia epidemic. Based on the existing intelligent manufacturing technology, a new type of intelligent mask translator with replaceable filter elements is designed. On the basis of satisfying personal protection, this mask can intelligently identify languages and translate, effectively solving the problem of language communication barriers in cross-border and cross-language communication.

The irony is a complex linguistic phenomenon used to express the opposite of its true meaning and is a more difficult part of sentiment analysis. Multimodal sentiment analysis complements each other with multiple modal data, thus reflecting the user's sentiment tendency more accurately. Sarcasm is often caused by semantic inconsistencies between modalities. In this paper, we focus on multimodal sarcasm semantic detection by analyzing the intra- and inter-modal inconsistencies. Firstly, image features, image attributes are extracted using ResNet, and then text features and label features are extracted using a pre-trained BERT model. The image features with image attribute splicing and text features are input into the self-attentive mechanism, and output inter-modal inconsistency. The text features and label features are input into the common attention mechanism and the output intra-modal inconsistency. After experiments sarcasm, semantic prediction accuracy can reach 86%.

Primary-level medical institutions are the first line of defense for the health of primary-level residents, and their information construction is an important means to promote the development of primary-level public health services and realize the modernization of medical services. The informatization of primary medical institutions has practical requirements for optimizing the medical treatment process, improving the service capabilities of high-quality medical resources, and reducing costs. This article focuses on the needs of information-based intelligent services in primary medical institutions, using new-generation artificial intelligence technologies such as speech recognition, text structuring, and experience interaction to build an electronic medical record collection system for primary medical institutions; based on the electronic medical record data collected by the system, primary medical institutions integrates the original basic public health service platform to open up various public health service platforms by using service-oriented institutional design, focuses on the intelligent services of medical record content verification, retrieval and data, and builds an artificial intelligence-based information service platform for primary medical institutions. It improves the quality of medical care, work efficiency and reduce operating costs fundamentally.

The VIKOR method was explored for multi-criteria optimization of complex systems. It determines the compromise ranking list and the compromise solution with the initial weight. ERP selection is part of key factors for companies to implement ERP. This paper puts forward the selection criteria of ERP software. Because the criteria value of an ERP software is fuzzy, the opinions of experts in the selection group are often inconsistent. In view of this situation, this paper puts forward ERP software selection of companies based on an extension of VIKOR method. By calculating the group utility, individual regret and benefit ratio with triangular numbers, and then rank the alternatives by the possibility matrix. The best ERP software is given. Finally, a case is illustrated to test the new method.

In the field of passenger cars, the weather resistance of vehicles affects the appearance, interior decoration and user experience of vehicles. At present, many vehicles have undergone automobile weather resistance tests, and a large amount of data has been accumulated. How to use big data analysis algorithms to get the vehicle temperature value, so as to quickly predict the temperature value of auto parts based on atmospheric environmental data, so as to obtain the environmental equivalent damage stress. Major automobile manufacturers and automobile weather resistance test institutions urgently need to solve problems. This paper proposes a vehicle weather resistance temperature prediction algorithm based on deep learning. By constructing a nested network that integrates long and short-term deep learning networks and full feature extraction networks. Nested network realize the prediction of the temperature of the vehicle measurement point, and obtain the predicted value of the vehicle measurement point temperature. The deep learning pre-diction model improves the efficiency of vehicle weather resistance analysis, shortens the weather resistance test cycle, and brings benefits to automobile manufacturers.

In order to discuss the error estimation of singular integral quadrature formulae, we construct new quadrature formulae with the anti-Gaussian quadrature, and the accuracy and remainder expression of quadrature formulae are given. The quadrature formulae are superior to the Gauss-Kronrod quadrature formulae in estimating the error.

Making full use of the historical data of the construction project and accurately and quickly predicting the project cost, can provide fast and reliable data support for the feasibility study, investment decision-making, quota design, scheme comparison and bidding decision-making of the project. Strengthening the accumulation and application of project cost data is not only the direction of the development of project cost information in China, but also the requirement of the State Council to promote the sustainable and healthy development of the construction industry. The principle and structure of deep confidence network are analyzed, and the confidence network model and RBM training model are discussed. Through the analysis of the change trend of project cost, the results show that the factor affecting the change of project cost may be the rise of steel price.

Given the dense distribution of points of interest in cities, it is difficult to semantically label trajectories by distance. This paper proposes a classification semantic labeling algorithm based on hidden Markov model (HMM). An algorithm based on time and space constraints is used to identify the stay points in the trajectory, and a density-based clustering algorithm is used to identify the user’s individual significant places. On the basis of the identification of individual points of interest according to the stay time and stay period, a method based on HMM is used to label public points of interest for the dense areas of points of interest. When calculating the observation probability matrix, a dynamic time function is used to simulate the influence radius of public points of interest in different periods. The simulation result shows that compared with the traditional method, the model has a significant improvement in the accuracy of labeling individual points of interest. At the same time, in areas with dense points of interest, the algorithm improves the accuracy of labeling by 6% compared with the traditional HMM.

Cancer has now become a prevalent disease that affects people’s daily life. Generally speaking, a woman has a 13% chance of suffering breast cancer at some point in her life, which means she has a 1 in 8 probability of developing breast cancer. Breast cancer is a form of cancer that early diagnosis is critical. Early detection of breast cancer can largely improve patients survival. As a result, several early detection or prediction methods are being developed in combating breast cancer. Computer-based cancer prediction has become a highly valuable tool, and it is widely used in the biomedical engineering filed. The combination of bioinformatics and data mining techniques are widely employed to aid in forecasting the likelihood of breast cancer incidence as well as cancer rates. In this paper, the aim is to discuss the fundamental ideas of bioinformatics and data mining, the mainly methods for cancer predication. In addition, the use of bioinformatics and data mining will be highlighted. This paper will provide the the comprehensive theory for the breast cancer treatment.

Due to the lack of illumination, uneven lighting, weather change, relative motion, overexposure and other complex natural environmental factors, the camera can not always capture high visibility images, which brings great challenges to the later application of machine vision. Therefore, it is necessary and imperative to post process the image to make it still have high-quality standards in a polymorphic environment. Low illumination color images not only have the characteristics of low brightness, but also are accompanied by noise and color offset due to equipment accuracy. In addition, it is easy to amplify noise, color offset, halo artifacts, unnatural gradient order and so on. For low illumination image, how to deal with its related problems, this paper introduces histogram equalization algorithm and Retinex algorithm.

In recent years, with the development of urbanization in all over the country, air quality has changed to varying degrees. Based on the air quality monitoring data, car ownership and industrial dust emission of 55 provinces and cities (except Hong Kong, Macao and Taiwan) in China from 2019 to 2020, Comparing various air pollution analysis methods, analyzing the characteristics of air pollution index in each city, finding an efficient and simple pollution control direction.

Infrared image has the low contrast and resolution, and there are few available features for infrared small targets, so the quality of bounding box obtained by the target detection model is substandard. To solve the problems, a model named CBAM-Yolov5 is proposed. Firstly, the convolutional block attention module is added to enhance the feature extraction ability of the backbone network. Then a scale bias factor is designed to improve the regression effect on the bounding box of small targets, which will increase the weight of the small targets on the loss function. Through experimental verification, the mAP and recall rate of our model can be 93.3% and 90.4% on the infrared ship dataset, and compared with Yolov5, it has increased by 1.1% and 2.0%, respectively.

The DBSCAN algorithm can discover clusters of arbitrary shapes, but it has difficulty in predicting the appropriate clustering parameters. In this study, the data field is introduced into the number field space, and the relative mass (RM) calculation method of the data field is proposed, and the first N points with larger mass in the dataset are calculated as the initial points of clustering by the RM algorithm. Then the optimized influence factor sigma is used to calculate the force range radius to achieve the optimization of the field radius parameter, so as to select the appropriate clustering parameters. In addition, this study improves the efficiency of computing large datasets by implementing the improved algorithm for parallel computing in a distributed cluster. Finally, the effectiveness of the improved algorithm is verified on three publicly available datasets, and the efficiency of parallel computation is verified on three large datasets. The results show that (1) the improved DBSCAN algorithm can effectively solve the problem of difficult selection of clustering parameters. (2) The maximum speedup ratio of parallel computation reaches 2.12 when the size of the large dataset is increased from 30,000 to 150,000 and the number of nodes involved in the computation is increased from one to five, and the average operation efficiency of the improved algorithm is improved by 32.45% compared with the original algorithm.

Screening the surface defect image of steel plate is a crucial link in the steel inspection system. In order to improve the detection efficiency, a defect image screening algorithm based on regional gradient projection is proposed. After the gradient map of the original image is obtained by an operator, it is divided into several equal-size regions which are projected horizontally and vertically, that can calculate the normalized standard deviation of each area in both directions, and then arrange in descending order among which the top 45% standard deviation is selected to the mean value to compare with the threshold to complete the defect image screening. The experimental results demonstrate that compared with the traditional gradient method, the accuracy of this method is improved by 8.95%, the missed detection rate is reduced by 5.33%, and the false detection rate is reduced by 3.62%. Therefore, the algorithm is simple, efficient and robust, which can meet the requirements of enterprise defect image screening

In recent years, image deraining has received extensive attention in image restoration and computer vision, and scholars mainly use it in fields such as video surveillance and autonomous driving. On rainy days, the rain marks will block the background of the scene, and the visibility of many scenes is low. In rainy images, the appearing alternately rain marks make it hard to differentiate between the background details and rain marks of the image, and the contrast and color of the target in the image will be attenuated to varying degrees, which leads to unclear expression of background information and makes some video or image systems unable to work properly. Therefore, it is very important to eliminate the influence of the rainy day on the scene of the image. This paper proposes the deraining network via the deep feedback based on the self-attention mechanism, the network uses feedback connections to feedback high-order information to low-order information and corrects the low-order information. Then inputs the image representation tensor fully processed by multiple feedback blocks into the self-attention mechanism, and the important features are fully extracted. The results prove that the network has strong reconstruction ability, and can generate clear and high-quality deraining images.

Fungi are of great functional significance in terrestrial ecosystems as the main decomposers. To better understand their decomposing process and population coexistence, we first describe and quantify the decomposition rate, focusing on three traits of interest selected by machine learning algorithm: moisture tolerance, hyper extension rate, and hyphal density and obtain, and use a ternary linear regression decomposition model (TLRDM) to quantify the decomposition rate. Then, to incorporate the interactions, we build an interactive decomposition model (IDM) and creatively employ a three-player logistic-based competition population model (TPLCM). Based on logistic growth, we formulate a differential equation group, fit the curves of this unsolvable equation group to obtain a function of population density versus time and compare the decomposition rates of three populations under interactive and non-interactive conditions, followed by analyzing the impact of the communications on decomposing ability. We obtain the population combinations that can coexist in certain climates. Furthermore, we include environmental factors, conducting a sensitivity analysis to describe how short-term and long-term climate changes affect our models.

In this paper, an expanded Decision-making Trial and Evaluation Laboratory (DEMATEL) approach is proposed to identify success factors. Second, initial qualitative evaluations are obtained by experts and quantitative evaluations are obtained by public data such as yearbook. Third, the quantitative matrix is constructed through calculate standard deviation and mean of the quantitative evaluation. Fourth, comprehensive direct matrix and total relation matrix are obtained respectively. Fifth, the centrality and causality are obtained. Finally, the case of Nanjing public health management capacity is illustrated to test the new method. The expanded DEMATEL method can also be used in other related fields.

With the development of modern information processing technology, the data collection of project cost information has become more and more complex, and the project cost management model is developing in the direction of networking and informationization. The original project cost information system can no longer meet the requirements of information communication and requires joint processing. Ant colony optimization algorithm has been widely used in engineering cost management. The ant colony optimization algorithm provides a practical solution for the entire life cycle of project cost data collection. This article first analyzes the cost data collection related to the entire life cycle of the power transmission and transformation project, studies the ant colony optimization algorithm, and the related characteristics of the mathematical measurement method, and proposes a series of data models. Analyze the main theories and characteristics of cost data. Through the research of related scholars in the field of "full life cycle engineering cost management", combined with the research results of predecessors and the author's academic achievements, certain suggestions are put forward for the collection of life cycle data.

This paper first uses the fuzzy comprehensive evaluation model to evaluate the impact of Saihanba on the environment quantitatively. Then, we determine the locations needed to build ecological protection zones in China. It is necessary to analyze the impact indicators such as energy consumption, the CO₂ emissions of each city, and forest coverage; and it can be easily solved through fuzzy comprehensive evaluation and clustering. Moreover, we also use the K-means clustering models to determine the ecological protection zones in need in Pakistan with the air quality index (AQI) of different cities. This will illustrate the generality and effectiveness of applying this environmental assessment method to different countries and cities in different countries. The final results show that the environmental analysis of each country or city by the method can effectively establish ecological protection areas in the corresponding regions.

Target tracking technology is an important research direction in computer vision, which has a wide range of applications, such as video surveillance, human-computer interaction, unmanned driving and so on. At present, target tracking technology has made great progress, especially the target tracking method using deep learning has achieved satisfactory results in the last two years, which has made a breakthrough in target tracking technology. This paper discusses the target tracking system from the two frameworks of correlation filter and Siamese network, and further examines the new target tracking method proposed. In the correlation filter, we analyze the improved algorithms such as CACF, ECO, MOSSE and SRDCF, However, all the above methods based on correlation filtering are affected by boundary effect. In order to overcome this problem, SRDCF came into being. SRDCF uses spatial regularization to punish the correlation filter coefficients, and obtains results comparable to deep learning tracking methods. In Siamese network, we analyze the algorithm frameworks of SiamR-CNN, SiamMask, SiamRN, SiamRPN++, SiamFC. In this paper, we review the object tracking task and split it into two main aspects: CF, Siamese network. For each framework, we survey the work and draw some opinions. Finally, we hope to help sort out ideas and help beginners get started quickly.

Parallel allocation is one of the most fundamental mechanisms for allocating indivisible objects to agents in a decentralized manner, in which agents are allowed to parallelly report their favorite objects among the remainder according to a policy that is insensitive to agent’s identities. In recent years, algorithmic issues about agent’s manipulations have been investigated, such as the computational complexity of verifying whether a manipulator can obtain a given bundle possibly, and maximizing her utility optimistically. In this paper, we consider the allocation policy of loser reporting, where the allocation process is divided into rounds, in each round, each agent that has obtained the smallest number of objects can report exactly one remaining object, and each reported object is allocated to one of the agents that report it at random. We show that for general additive utilities, under the optimistic assumption (i.e., the manipulator can obtain an object once she reports it), an optimal manipulation can be computed in polynomial time with respect to the number of objects.

Pattern knowledge plays an important role in text recognition. The classification and segmentation system under a large concept system needs to focus on the problem of pattern knowledge mining. The relationship between labels and concepts is divided into three types: equal, belong, and irrelevant. There is a need for a classification method to accurately classify the relationships between label and concept, and between concepts. In this paper, the classification algorithm proposed is SA-KNN-SVM, which can retain the advantages of fast K-nearest neighbor (KNN) model training time and good prediction effect while improving accuracy. The SA-KNN algorithm aims to continuously adjust the parameters and determine the number of the iterations through loop iterations, and then quickly find the accuracy rate corresponding to the different K values. SVM has many kinds of kernel functions, select the kernel function that makes the classification accuracy higher. And find the best parameters by looping an iterative grid search. The experimental results of the proposed algorithm show significant improvement in classification accuracy and processing time.

Diabetes is a chronic disease characterized by hyperglycemia. According to the statistics of estimated the number of people with diabetes reached 537 million in 2021. The rapid increase in the number of diabetics makes diabetes become a global health disease threatening mankind. Diabetes is usually accompanied by a long period of diagnosis. Before diabetes is discovered, the body is in a state of high blood sugar for a long time, which will cause serious complications for patients, early detection and treatment of diabetes can greatly alleviate the harm caused by the disease. In this context, this paper proposes a Diabetes prediction algorithm model based on PIMA Indians Diabetes Dataset(PID) published by the University of California at Irvine. In terms of model architecture, the prediction algorithm model is stored in the cloud server center to give full play to the efficient computing power of the cloud server, and it is more convenient to iteratively update the model algorithm. The algorithm model is divided into two parts. The first part uses the neural network to raise the dimension of the data. Inspired by the Cover theorem, the nonlinear dimension of the data is easier to distinguish the data. In the process of raising dimension, the loss function is used to make the data separability after raising dimension stronger. The second part uses the raised dimension data to train the base classification algorithm and integrate the trained algorithms with stacking methods to obtain the final model. Finally, the prediction accuracy of the algorithm model is 89.72% on the test set, which is significantly improved compared with other algorithms proposed on the PID data set.

Taking Zunyi conference site as an example, this paper studies the spatial and temporal distribution characteristics of tourism network attention of Zunyi conference site in 2019 by using seasonal intensity index. The findings are as follows: 1.) In 2019, the network attention of Zunyi Conference site generally showed an upward trend, with the largest increase in June-August, and the largest increase in the Spring Festival, May Day and November Day during holidays. 2.) There are obvious spatial and regional differences in the attention of Zunyi Conference site network, mainly distributed in southwest China, especially in Guizhou Province and surrounding provinces and cities. 3.) Most tourists who pay attention to Zunyi Conference site are young tourists aged 20-39, and male tourists pay more attention than female tourists.

Traffic flow prediction is of great significance to the development of intelligent transportation. Since most of the current research is to improve the prediction structure, ignoring the impact of the nonlinear and unstable characteristics of the data itself on the prediction, this paper proposes an algorithm that combines data decomposition and spatiotemporal correlation prediction. In the decomposition part, VMD (variational mode decomposition) and CEEMDAN (complete ensemble empirical mode decomposition with adaptive noise) are introduced to decompose the original data, and the subsequences are adaptively aggregated based on the permutation entropy theory. The spatiotemporal correlation prediction part uses DTW (dynamic time warping) to select the site most relevant to the measured site, and then weighs the predicted values of multiple sites according to the rank index method to obtain the final predicted value. Finally, the experimental results show that the method in this paper can extract traffic flow trend information well, and achieve better accuracy in the short-term forecasting.

Aiming at the magnetic research of unexploded ordnance with various attitudes and shapes, a magnetic measurement and analysis method of unexploded ordnance is proposed. In this paper, the model of unexploded ordnance magnetic field source is established, and the equipment for multi detector parallel acquisition of unexploded ordnance magnetic field data is designed. After obtaining the information of unexploded ordnance magnetic field, the data repair and inversion calculation are carried out, so as to obtain the magnetic characteristics of unexploded ordnance. The experimental results show that this method can not only quickly and effectively obtain the magnetic field information of the detection surface of unexploded ordnance, but also retrieve its magnetic moment and buried depth, and the error is less than 20%. This research has important reference value for the magnetic analysis and detection of unexploded ordnance.

GAN, proposed in 2014, has received high attention and extensive research from industry and academia. It is an important node in the development of generative model. With the development of research, GAN has been improved in theory and application. The basic idea of GAN is derived from two-person zero-sum game, which consists of a generator and discriminator, and is iteratively trained through confrontation. The purpose is to learn the distribution of sample data and generate new samples. Firstly, the theoretical framework of GAN is introduced. Secondly, some typical GAN models are discussed. Then four applications of GAN in data generation, image super resolution and so on is shown. Then the advantages and disadvantages of GAN are discussed and the future development direction of GAN is pointed out. Finally, the future prospect is summarized.

Facial expression recognition is a hot research topic in artificial intelligence industry and has a good research prospect in various fields. At present, facial expression recognition processes the whole face directly, but the pixel value of non-feature region may bring some interference for feature descriptor extraction. Considering that the cartoon effect of the face can directly reflect the facial expression features. In order to make the network pay more attention to the information of the facial features and their surrounding pixels, this paper proposes an expression recognition algorithm based on key point detection of Covering multi-scale Practical Landmark Detector (CM-PFLD). Under this algorithm, this paper constructs a cartoon expression data set, which only retains the key points of facial expression information, and then classifies facial expression by directly locating the key points of facial expression information. In order to verify the feasibility of the expression recognition method in this paper. The experiment uses Fer2013 and CK data sets to produce cartoon expression data sets, and trains and compares cartoon data sets and original data sets respectively under the same network. The experimental results show that the method proposed in this paper has high detection accuracy and fast speed on standardized and neat data sets. On the data set with more unfavorable factors, the training accuracy of the two methods is similar, but the processing speed of the proposed method is faster. Experimental results show that the proposed method is feasible and effective.

Tourism has become one of China's important economic pillar industries, and the revitalization of the tourism industry in the context of the COVID-19 epidemic is an issue worthy of attention. This article will take Guilin network attention as the research object, Baidu index data as the data source, seasonal concentration index and geographic concentration index as the research methods, and use Python software to visualize Baidu index data. Through the visualization of the monthly, weekly and geographic spatial distribution of the data, it can be seen that: 1. Guilin's network attention under the background of the COVID-19 epidemic is sluggish; 2. Guilin's annual tourism network attention has seasonal time concentration and geographic spatial distribution. This is to predict the situation of tourism demand during the post-epidemic period in Guilin in the future, and to prepare for the recovery of tourism after the epidemic in Guilin.

In open-domain question answering, especially in multi-hop question answering, it is important to extract clues related to the question from the huge knowledge base for the next-hop reasoning. Most of the current question answering methods do not make full use of the knowledge base to retrieve clues. To answer complex questions more accurately, in this article, we first extract the clue paragraphs related to the problem from the Wikipedia knowledge base to construct the representation of a fine-grained node. And then we perform multi-hop inference on the structure graph based on these nodes. An elaborate evaluation was conducted based on two public data sets. The proposed model can extract more answers and clues related to the question by importing an external knowledge base and achieves higher accuracy than the benchmark method on multi-hop reasoning tasks in joint answer predicting tasks.

This paper takes typical air defense equipment as an object, analyzes the characteristics of the data in the whole life cycle of equipment, sorts out the data mining requirements in support design of equipment in demonstration, scheme and development stages, including spare parts design analysis, fault analysis, preventive maintenance analysis, state-based maintenance analysis, using data analysis requirements involved in data mining, and puts forward the realizing methods of data analysis and mining. It can greatly improve the comprehensive support design level of air defense equipment.

By optimizing the process parameters, we can reduce the die trial time, lower the production cost and improve the forming quality of the stamped parts. Taking cistern forming as an example, a computer aided engineering model was established to reduce the maximum thinning rate. The back propagation (BP) neural network was trained by combining the partial factor test method with the numerical simulation of stamping process, using the friction coefficient, die clearance, sheet thickness, binder force and the drag coefficient of the draw bead obtained from the simulation experiments as input values and the maximum thinning rate as output values. Particle swarm optimization (PSO) was used to optimize the stamping process parameters. The conclusion has guiding significance for stamping process design.

In the whole lunar regolith drilling and sampling task, It is critical to make the operation of the drilling and sampling system efficient and reliable. In view of the high efficiency drilling requirements of lunar rock under low-force conditions, it is necessary to study the mechanical properties of an elemental blade and lunar rock. Firstly, a test platform is developed to test the low-force cutting load characteristics for the lunar rock simulant according to the requirements of the cutting load characteristics test function. Then, the linear cutting load and impact crushing load characteristics for the lunar rock simulant are tested and verified respectively. It is finally found that the cutting load of the lunar rock simulant can be reduced by increasing the cutting edge rake angle and side rake angle and improving the percussive frequency. This provides a certain reference for setting the operating parameters of lunar rock core drilling tools to achieve high performance drilling under low-force conditions.

The national economy has also brought about the change of the payment method. The credit card has changed from a convenient payment and settlement tool to a popular consumer credit tool. The consumption data of credit card is huge, it is difficult to identify whether it is in default by manual, and the traditional statistical analysis technology is also difficult to solve such a complex task. Therefore, this study uses data mining methods, for the data pre-processing, data statistics and analysis, data visualization and data analysis report writing as the core, and based on the specific analysis of the current situation of China's credit card business, carries out data cleaning from three aspects: Missing value processing, Abnormal value screening and dataset division. It also makes exploratory analysis combined with descriptive statistics and data visualization, so as to further establish the credit card default identification model. And put forward suggestions, improve customer information, strengthen information sharing within the industry, strengthen risk data analysis, rational formulation of risk prevention and control strategy, and future research operation skills. Finally, it draws a conclusion and puts forward the prospect of operation skills for future research.

In this paper we propose a generalized regression neural network (GRNN) algorithm to rapidly identify adulterants of milk powder. Milk, milk powder and other dairy products are rich in nutrients such as protein, carbohydrates, and fats. The demand for dairy products are increasing persistently. However, incidents of adulteration of dairy products occurs a lot. To obtain huge profits, many unscrupulous merchants add melamine, starch and sugar instead of nutrients to the milk powder, leading to the mendaciously increasing content of protein in the powder. Conventional dairy product content detection methods, which have devices with large weight and volume, need to take a long time to measure the content of adulteration, and thus they are not suitable for rapid on-site detection. Therefore, we use a GRNN model to rapidly identify the adulteration by measuring the Raman spectra of adulterated substances in dairy products. The GRNN model can be trained for classifying the Raman spectra of adulterated substances. The result shows that the model can achieve 100% accuracy in identifying milk powder, melamine, wheat flour and corn flour, and the identification time of algorithm is seconds-level, laying the foundation for the development of rapid detection technology for adulterated dairy products.

Based on the intensity distribution of continuous atomized rain, a coupled mathematical model of slope runoff and seepage and its numerical simulation method were established. According to the test results, the continuous distribution of atomized rain intensity corresponding to the elevation on the slope is obtained; the one-dimensional kinematic wave model is used as the slope runoff equation to describe the slope runoff of atomized rain acting on the slope, and the flow velocity of slope runoff is expressed by the Manning-Strickler formula; the saturated-unsaturated seepage equation is used to describe the groundwater seepage in the slope when atomized rain infiltrates into the slope. Unified boundary conditions are used to couple the runoff field and seepage field in the slope for the flow velocity and water head on the slope. The Galerkin finite element method is used to numerically simulate the slope seepage and slope runoff process when atomized rain affects the bank slope. The method is applied to a practical engineering slope, and the results show that the method can analyze the slope infiltration under the condition of flood discharge atomized rain, and can accurately simulate the slope infiltration process of atomized rain coupled with runoff and seepage.

Under the background of the new era, technologies such as big data and cloud computing have gradually penetrated into the power industry, providing strong support for data collection and quality improvement of power energy. In this regard, this article uses a distributed architecture of electricity consumption information collection system to collect electricity consumption data and establish a big data cloud platform. On this basis, through BP neural network algorithm and other big data analysis methods, accurate load forecasting and environmental pollution prevention and control are implemented. , To improve the effectiveness of line loss management, and reasonable prospects for the future development of power energy big data, hoping to better play the value and role of power energy big data, and support the long-term stable development of the power industry to provide a certain reference and reference.

In the precision engineering measurement, the total station is often used for free station measurement, and improving the accuracy of stitching the measurement data of each station is the premise of widely used total station measurement by free setting. This paper is based on the quadratic method to study the stitching method of data measured by total station with free set-up stations. The total station data is obtained by free setting based on the quaternion method, and the method of stitching the data is investigated. Taking the deformation monitoring of the wooden tower in Yingxian County as a case study, a total station was measured by free setting, and the quadratic algorithm was used to realize the coordinate conversion between stations by point name matching, which is comparable to the stitching results of Leica's Cyclone software. In this paper, the improved quaternion algorithm is applied to the free station splicing of total station with the features of fast matching, coarse difference detection and rejection, optimization of redundant observation of the same name point splicing algorithm, inclusion of splicing error correction, and high splicing accuracy.