This PDF file contains the front matter associated with SPIE Proceedings Volume 11933, including the Title Page, Copyright information, and Table of Contents.

Due to the lack of label information for new items, the recommendation effect of traditional recommendation algorithms will be reduced. To solve this problem, this paper proposes a recommendation algorithm based on Word2Vec and convolutional neural network. First extract the keywords of the item text description information, use the Word2Vec model to convert the keywords into word vectors, calculate the similarity matrix of each keyword, and then use the similarity matrix as the input layer of the convolutional neural network to obtain the similarity between the items. Calculate the prediction score of the new item, and finally consider the user's preference for attribute information to generate recommendations. The experimental results show that the algorithm has a good effect on the hit rate.

The propose of this paper is to develop and study an artificial intelligence-based optimization algorithm for the famous “Flappy Bird”, which benefits to a better game score. We first create an agent that learns how to optimally play the game by safely dodging all the barriers and flapping its wings through them. In order to expand the feasible solution space and generalization ability, a multi-gene genetic algorithm based on the Genetic Algorithm and Neural Network is proposed. We carry out detailed experiments to prove the proposed multi-gene genetic algorithm to be useful and efficient. The quantitative results show that the multi-gene genetic can significantly improve the game score.

With the rapid development of carbon neutral concept, various car companies have gradually listed it as the main research object, which is of great significance for the development of the automobile industry. At the same time, although electric vehicles have no carbon emissions in use, they still have large carbon emissions throughout their life cycle. In this paper, in the context of the carbon neutralization goal of the world, we study the actual carbon emission control of electric vehicles, explore a method of carbon emission management and prediction of electric vehicles based on BP neural network learning algorithm, and realizes prediction before volume production and analysis of carbon emissions while the car in use. Through the construction of neural network and prediction model, the precise management of electric emissions can provide significant reference to existing car companies.

Rolling bearing is an important part of rotating machinery. On-line condition monitoring and fault diagnosis of rolling bearing are of great significance to ensure the stable operation of rotating machinery. In this paper, a novel rolling bearing fault diagnosis method is proposed based on continuous wavelet transform (CWT) and transfer convolutional neural network. Firstly, the vibration signal is converted into time-frequency image through CWT. Then, on the basis of the pre-trained convolutional neural network model, the construction of rolling bearing fault diagnosis model is conducted based on transfer learning. Moreover, an open source data set obtained from the Bearing Data Center of Case Western Reserve University was used to verify the effectiveness of the proposed method. Finally, the comparative analyses were conducted among the proposed method and other rolling bearing fault diagnosis method. The results show that the proposed method can achieve higher accuracy on the accurate rolling bearing diagnosis of fault modes and fault degrees.

In order to detect the electrical fire hazards of distribution boxes in heritage buildings and achieve the purpose of electrical fire risk prevention, this paper has established a multi-sensing pyrolytic particle electrical fire early warning analysis model based on BP neural network, and carried out the experimental verification according to the national standard GB 14287.5 “Electrical Fire Monitoring System Part 5: Measurement Pyrolytic Particle Electrical Fire Monitor Detector”. Firstly, according to the features and rules of high temperature pyrolysis of various electrical insulating materials in distribution boxes, this paper analyzed correlation between different electrical insulating materials or wood surface temperature and risk of electrical fire hazards, and a risk analysis algorithm is established to quantify electrical fire hazards. Then, it established a BP neural network model based on the mass concentration of pyrolytic particles and VOC gas concentration. Finally, through the pyrolysis experiments of various electrical insulating materials and wood, this paper conducted data analysis and verified the algorithm model. The experimental results show that the established BP neural network analysis model is effective with mean absolute percentage error of risk prediction of about 11%. The fitting result of this model is good, and it can be applied to pyrolytic particle electrical fire detection of distribution boxes in heritage buildings.

The evaluation indicators of wireless sensor networks include physical layer and data layer indicators. Although physical layer parameters such as received signal strength indication (RSSI) and signal-to-noise ratio (SNR) can reflect link quality to a certain extent, the stability of wireless sensor networks will eventually be reflected in the data layer, namely packet reception rate (PRR). If the data layer evaluation parameters are used to monitor and evaluate the link quality of the LoRa (Long Range) wireless network, this method will not have real-time performance, and the real-time channel quality evaluation is for the nodes in the adaptive frequency hopping network to accurately access the wireless channel a key technology. Aiming at the phenomenon of unreasonable channel resource allocation in wireless networks, the Elman neural network algorithm is selected as a tool, combined with the physical layer parameters RSSI, SNR and data layer parameters PRR that can be easily obtained from the chip, and the mapping relationship between the three is trained. Real-time channel quality assessment, improves the reasonable allocation of channel resources and the reliability of communication quality.

Through the investigation and evaluation of the enterprise's scientific research capability, the important indexes affecting the enterprise's scientific research capability are analyzed and summarized. Combined with the associative memory capability of the discrete Hopfield neural network, the evaluation model of the enterprise's scientific research capability based on discrete Hopfield neural network is established, and the ideal level evaluation index and code are designed. The discrete Hopfield neural network was created by simulation software. By taking the code of the enterprise level evaluation index’s score to be classified as the input of discrete Hopfield neural network, the scientific research capability of the enterprises can be evaluated objectively, fairly, quickly and accurately after training.

This paper proposes a snake scanning focus-spot reconstruction method based on multi-pixel information fusion, which can detect the centroid-position and relative-intensity distribution of small-scale focus-spots. The theoretical analysis is consistent with the experimental results.

Shortest path algorithm is the basis of many optimization problems such as resource allocation and optimal path planning. It is very meaningful to study the shortest path algorithm, because it can reflect the efficiency. We propose an improved Dijkstra algorithm based on a new storage method, which improves the shortcomings of the Dijkstra algorithm to a certain extent by constructing the map of the forward star by the chain. The experimental results show that our improved algorithm is more effective than the classical algorithm. We provide an effective method for the analysis and application of actual large graph theory problems.

Chinese text classification has been in the research stage, there are many machine learning algorithms that can be used, such as logical regression, SVM, KNN, naive Bayes, random forest, neural network and so on. In this paper, taking Chinese modern novels as an example, we use various algorithms for classification and comparison, and choose the best algorithm for naive Bayes and neural network. After adjusting the TF-IDF algorithm and processing the participle according to the TF-IDF value, the accuracy of classification is improved obviously. The logistic regression with the lowest accuracy can increase about 6.7%,while the simple Bias and neural network can reach 100%.

The rapid development of image recognition technology has tremendously promoted the research of species detection. Detection of rare species has important implications for protecting biodiversity. This paper proposes a deep learning method to solve bird species detection based on the current powerful algorithm YOLO-v5. The algorithm combines the feature extraction network of CSPDarkNet to extract the feature information of the input bird image and then uses the aggregation network with a pyramid structure for detection to obtain the required semantic information. The experimental dataset comes from the Kaggle platform and is annotated with free software provided by Github. In this experiment, YOLO-v5 and the other five models are compared for the performance of bird species detection. The comparison results shown that YOLO-v5x has the best precision (95.36%) and mAP (91.28%). It can be seen that compared with other models, YOLO-v5x has high recognition accuracy, good robustness and can be better suited in the actual environment.

Network analysis can help discover latent information on graph-structured data, and one of the major topics in link prediction [1]. Given the current state of a graph, the main task of link prediction is to predict the emergence of currently non-existing associations between nodes [2]. This project handles this problem with a machine learning-based method, where we compute a series of network parameters, feed them to a simple neural network and obtain the likelihood label for each non-existing edge. In the topic of connection prediction, this paper focuses on the application effect of mainstream machine learning algorithms, finds the key statistical parameters in the algorithm, and compares the prediction efficiency with deep learning. The test data I used came from Facebook. A complex network with 4000 nodes and 88000 links is constructed as the test object, and the influence of test sample data selection on test results is analyzed.

With the vigorous development of the modern logistics industry, traditional route optimization and matching can no longer meet the increasingly complex road conditions and changing natural factors. Being aware of the low efficiency and high energy consumption problems of vehicles in logistics distribution, we propose a logistics distribution route optimization method based on data mining, combined with mathematical models, calculates the model based on the number of samples, combined with GIS technology for accessibility analysis, and compares the distribution speed, logistics cost and economic benefits of different routes, in order to solve the complex optimization problem of logistics path in different environments.

In order to optimize the decision-making process of the traditional auxiliary decision-making model, the author used the feature of big data mining can extract or excavate hidden valuable information from complex and multi-scale massive data, designed a big data decision support model which is based on deep learning algorithm and improved the original “six libraries” decision model. Its principle and operation process were described deeply. The process provided intelligent and informational auxiliary methods for commander’s decision making, and improved the real-time and accuracy of Command and Control (CC) decision making. Furthermore, the further applications of data mining and deep learning algorithm in CC Information System were discussed.

Air pollution has been plaguing human society and seriously affects the normal activities of human beings. PM2.5 is a kind of air suspension particles with complex composition, which is one of the main components of air pollution and poses serious threat to human health. In this paper, a model based Temporal Convolution Network (TCN) was applied to predict outdoor PM2.5 in Beijing. Model based on TCN is capable to parallel operations and has flexible convolution kernel compared with models based on other neural networks such as Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM), Gate Recurrent Unit (GRU) and traditional Convolutional Neural Network (CNN). Additionally, the model was verified in the experiment and the results indicated that the model achieved better performance than traditional models.

In account of the context of big data, it is a trend for the airport to build “intelligent security” mode, which is conducive to the control of airport security work. By applying the core technology of calculator, each feature of the suspect can be predicted with the mode. The most suspicious person will be predicted through the calculation of the similarities between the predicted features and the personal information of the passengers who will take the flight. Based on the XLMINER system, through the discriminant analysis, K-Nearest Neighbor (KNN) proximity algorithm, classification trees, Naive Bayesian algorithm analysis within the classification tools, the experimental results showed that KNN proximity algorithm has better predictive ability for predicting the suspects.

Shared bicycle is an emerging industry in recent years. It is an important part of urban transportation system. Its shortterm demand forecasting is of great significance to the supply, management and allocation of shared bicycle resources. The data of shared bikes are crawled to analyse the impact of time and climate characteristics on the demand for shared bikes. The short-term demand of shared bicycles is predicted by long short-term memory neural network. The experimental results showed that the long short-term memory neural network is suitable for the prediction of shared bicycle demand, and the prediction results with climate characteristics are better than those with only time series. Applying this model to predict the short-term demand of shared bicycles can improve the configuration efficiency of shared bicycles. On this basis, it provides a basis for establishing accurate and effective shared bicycle configuration strategy.

Humor prediction based on the statistical architecture aims to automatically judge whether a sentence/dialogue is laughing-provoking or not. Due to the complexity of the humor prediction, it is necessary to consider the contextual relevance on the text data when conducting semantic analysis, and the accuracy of Humor detection or comparison need to be further improved. This paper presents to the usage of Bi-directional Long-Short Term Memory (BiLSTM) to predict humor on textual data, and the predictions are reflected in the form of the comparison of humor between the two texts. In order to compare the two sentences in a mathematic way, the sentences are projected into vector space and the sentences are represented at the word-level granularity by using Global Vectors for Word Representation. To strengthen the influential factors of specific words in humor prediction, the similarity of humor vector (SoHV) is introduced as an additional feature, which is calculated from the Cosine similarity between the edit words and the original words in sentences. BiLSTM is compared with Convolutional Neural Network (CNN) on Humicroedit to illustrate its effectiveness in dealing with textual data. The experimental results illustrate that the proposed method significantly outperforms CNN by obtaining the accuracy of 51% and 34.2%, respectively. The effectiveness of BiLSTM is further verified on SemEval-2020 Task 7, which assesses Humor in Edited News Headlines. We are ranked 46^th and 64^th as to subtask-1 (classification task) and subtask-2 (regression task). Our analyses show that some particular words do have huge impact on the degree of humor in whole sentences and the BiLSTM model can have better cognition capacity by utilizing SoHV.

In the recent years, the unmanned combat aerial vehicle (UCAV) techniques is a hot topic of research. Many researches are studying how to use to fulfill missions and defend enemies based on simulation platforms. Different AI agents have been constructed to control virtual UCAVs to perform tasks on simulation platforms. Rule based AI heavily depends on human knowledge and lacks of flexibility. They cannot adapt to the changing environment. Reinforcement learning based AI has advantages over rule based AI as its depend less on human knowledge. In this paper a hierarchical reinforcement learning method is proposed on Multi-UCAV air combat based on simulation platform. The experiment results showed that the hierarchical approach can outperform state-of-the-art air combat method.

Humor detection is one of the most popular tasks in natural language processing. Yet humor is abstract to numerate, and there isn’t an acknowledged standard for humor assessment. Toward this end, this paper explores the performance of CNN, RNN, BiLSTM in tackling humor detection. We use the regression method and classification method, respectively, to identify the best model. The experiment was conducted on a dataset that is composed of 15,000 news headlines. Results show that the CNN network is a preferable choice, and combined with the classification method, the model obtains the best performance. Though there is plenty of more sophisticated sentiment analysis models, our work offers an intuition for short text sentiment analysis.

A large number of criminal investigation images are collected for modern case detection, and these images not only contain many valuable clues, but also can provide powerful evidence. At present, most of the current criminal investigation image retrieval methods for public security investigation applications use text-based or traditional shallow feature-based image retrieval methods, and the accuracy and efficiency of retrieval can hardly meet the needs of modern criminal investigation cases. In this paper, we summarize and analyze the current research results and technologies in the field of image retrieval, and adopt the image retrieval method based on depth features to improve the accuracy and efficiency of the criminal investigation present investigation image retrieval. In the depth feature-based image retrieval, VGGNet and Res Net are fine-tuned using the criminal investigation image dataset, and then the image depth features are extracted for retrieval experiments. The experimental results show that the retrieval model has the following two shortcomings: the model cannot adapt to the target scale change; the retrieval accuracy is lower than the average in the category with fewer samples. Two optimizations are proposed to address these problems: introducing pyramid pooling to improve the robustness of the model to target scale changes; retraining the network after enhancing the data samples to make the retrieval accuracy of the model more balanced for different categories of samples. In addition, query expansion is introduced to enhance the image feature representation during retrieval. After using the above optimization methods, the retrieval accuracy is improved by 5.7%.

In view of the problem of complex feature engineering and poor learning ability in network traffic anomaly detection, this paper proposes an improved one-dimensional convolutional neural network(1D-CNN), which is mainly composed of a set of parallel convolutional layers, a global average pooling layer and a dropout layer. First, a parallel structure and an increasing number of convolution kernels are utilized to capture features of different scales. After feature fusion, global average pooling is used for information aggregation. Finally, the fully connected layer is applied for classification. Experiments on the model obtained by means of parameter tuning and training show that the improved 1D-CNN enhances the ability to learn features and reduces the information loss of the pooling zone between convolutional layers. The average precision reaches 95% and overfitting is suppressed. Compared with other machine learning algorithms, the proposed model has better performance in abnormal network traffic detection.

In order to meet the needs of the economic market for high-level talents and solve the deficiencies in education and training, this paper carries out the exploration of higher vocational education and training reform based on BP neural network, analyzes the advantages and disadvantages of the application of BP neural network in education and training reform, and locates the deficiencies and reasons in Higher Vocational Education and training. Based on the cloud platform, improve the reform measures of higher vocational education training curriculum system, and deeply integrate the training contents obtained from multiple channels, so as to improve the adaptability of education and training contents to industry policies. The BP neural network is introduced, the sigmoid function is used to activate the BP neural network nodes, analyze the higher vocational education training data, and calculate the higher vocational education training evaluation error. According to the education and training evaluation mechanism, BP neural network is used to construct the higher vocational education training evaluation model to evaluate the teaching results in different stages online. The experimental results show that the evaluation accuracy of this method is 91.65% and the evaluation time is 2.09 s. Based on the evaluation results, the macro adjustment of teaching and training progress and objectives is carried out to ensure that the adjusted training objectives and tasks can meet the basic needs of the educated groups.

As legal case judgment documents begin to be published on the Internet, legal data is increasing. At the same time, with the continuous improvement of computer computing power, the continuous development of data mining, machine learning, deep learning and other technologies, the era of legal big data must Will come. This article has carried out in-depth excavation of judgment documents of legal cases, in order to provide an objective and fair reference for the sentencing of legal cases, and proposed a legal case decision-making system based on neural network. This article first extracts information in accordance with certain regular rules from the judgment documents of legal cases, and uses the random forest model to predict the sentences in the judgment documents. According to the specific details of the case, determine the existence of various sentencing circumstances in the case, and use it as a characteristic quantity for classification and decision-making. In order to improve the decision-making performance of the network, in view of the local minima and false saturation existing in gradient learning, a learning algorithm based on neural network is proposed, which further improves the decision-making ability of the network. Simulation experiments show that the decision-making network can give an objective and fair sentencing result.

Humor is a high-level semantic emotion that can only be understood at a stage when the human mind has developed. Humor detection is a challenging task in the field of natural language processing. In this paper, we focus on the characteristics of humor from the way it is generated and propose the Humor Important Message Attention Net (HIMA-Net): a self-attention network based on the key messages related to humor. Results show that HIMA-Net outperforms the traditional models on three datasets (Headline, Pun, Short Jokes), and further analysis demonstrates the effectiveness of the proposed model.

The development of weapon equipment construction is a complex and dynamic giant system, which involves many aspects, such as development strategic planning, formulation of policies and regulations, establishment of management organization, establishment of operation mechanism, allocation and utilization of resources, and adjustment of industrial structure. This paper introduces the modeling purpose and idea of system dynamics. Starting from the basic structure of weapon equipment construction and development, it analyzes the relationship between various elements in the system and between the system and the environment by using the theory and method of system dynamics, establishes the weapon equipment construction and development model based on system dynamics, and simulates the development process of weapon equipment construction. The research results provide a reference for deeply grasping the development law of weapon equipment construction, and provide a new idea for scientific and reasonable planning of the realization path of weapon equipment construction.

As PBL (Project-based learning) mode has been widely and intensively integrated into many curriculums of higher vocational education, a lot of research has been done in order to assess the impact of project-based learning on higher vocational students’ English learning process. Yet, most measures are relatively subjective and lack in accuracy. This article aims to propose the method of combining AHP and BP neutral Network to construct the evaluation model of project-based English learning. AHP is used to comprehensively consider the selection of evaluation indicators from various factors affecting learning effectiveness based on PBL English learning and to set the weight of each evaluation index for PBL learning according to a questionnaire and experts’ scores. Then applying the calculation data of the AHP method as the target value of the BP neural network training and thus a BP neutral network-based evaluation model of PBL English learning is built. The simulation experiment denotes improvement on the objectivity, accuracy and efficiency of the evaluation on learning effectiveness of PBL English learning.

One of the biggest challenges in High Resolution Remote-Sensing Interpretation based on deep learning technology is the access of huge number of sample annotations. Currently, the sample annotation is carried out manually, which is a boring, time-consuming and labor-consuming work. Remote-Sensing Imagery comes from a variety of sources, each with its own independent features, so that there is insufficient quantity of sample annotations to cover most of these features. The Third National Land Survey (TNLS) data is a national land survey with high-quality manual annotation. This paper discusses the use of TNLS data instead of manual sample labeling for model training, and three hybrid methods of sample cleansing, data augmentation and training in turn, to create a deep learning training method of water identification. The results shown that the TNLS data with three hybrid methods was able to identify high-quality waters, the Frequency Weighted Intersection over Union (FWIOU) was up to 91.8%, which was 5.9% higher than the conventional training methods. The deep learning model used in this research was proposed to obtain the high accuracy and saved a lot of human resources, which had a wide range of practicability.

Digital media technology is widely used in all walks of life, research digital media information sharing platforms in the Internet era. In the hardware design of the platform, hicd-3480 hybrid architecture intelligent communication device is adopted. In terms of software design, based on the Internet era, we integrate heterogeneous distributed database, further set various functional modules of the platform, design the download page of the sharing platform, and complete the research of digital media information sharing platform in the Internet era. The experimental results show that the functions of the design platform work normally, and the six preset test objects are obtained. The design platform is in normal working state, which provides an effective reference for the digital media information sharing platform.

How to use the limited attack resources to attack the battlefield communication network is an essential problem in network countermeasure. This paper proposed the concept of weighted separation Ratio (WSR) of the network topology and corresponding optimal attack strategy based on WSR was examined. Firstly, the equivalent weighted topological model of the network is established, and the concept of WSR is proposed to describe the degree of network separation. And then, considering the attack cost, the optimization mathematical model of network attack strategy is developed. Finally, the correctness and effectiveness of the method are verified by theoretical proof and example analysis.

Predicting the duration of traffic accidents can effectively help traffic management. To make a more accurate real-time prediction of traffic accident duration, and fully utilize the huge amount of traffic texts in social networks, in this paper, we consider this prediction task as a classification problem. First, the reported text of traffic accidents in social networks is obtained. After the data augmentation, the Bag-of-words model and Fisher optimal segmentation algorithm are combined to calculate the optimal classification threshold based on duration, and the accidents are classified into four classes. And then, the C-BiLSTM neural network is constructed by fusing convolutional neural network (CNN) and bidirectional long short term memory (Bi-LSTM) to predict the classes of accident durations, and the prediction accuracy of final trained model can reach 96.09%. Through experiments, the proposed method is proved to be practical and effective in solving traffic accident duration prediction.

Target detection based on deep learning methods has important applications in the field of remote sensing images target detection. Aiming at the disadvantages of the small target scale for remote sensing images, this paper proposes an improved YOLOv4 algorithm to improve the detection effect of remote sensing images. The experimental results show that the algorithm proposed in this paper has an average accuracy rate of 2.25% higher than that of the original YOLOv4 algorithm. The accuracy of detecting small-scale targets has been significantly improved, and the amount of model parameters has also been reduced compared to the original algorithm.

This paper introduces a wearable human motion capture system based on inertial sensor. The sensor nodes are connected by network cables, while the hub node transits data with the computer through Bluetooth. This system not only ensures the data acquisition rate, but also has no movement restrictions for users, such as requiring them to move within a certain range. Using Python programming and Blende engine, the inertial data of each part of the human body is converted into rotation angle and displacement to drive the human model. Meanwhile, rich human-computer interaction functions and visual interface are designed to reduce the difficulty of use. The final motion capture results show that this system has a good performance in recognizing complex human movements, and can be used as a multi-functional motion capture system in motion analysis, film and television production, virtual reality and other fields.

Laser-EMAT (electromagnetic acoustic transducer) technology has the advantages of laser induced ultrasound and EMAT simultaneously. This paper introduced a novel simulation about laser-EMAT testing the surface crack with various depth in the aluminum block. A finite element model of laser-EMAT detection was set up for investigating the influence of groove-type crack on ultrasonic propagation in the way of numerical simulation. Then, the response curve of voltage in time domain was obtained by the testing coil in EMAT, which is proposed to determine the depth of crack above. Good coupling could be found between voltage of signal received by EMAT coil and amplitude of ultrasound generated by the laser. In addition, the snapshot of ultrasound field at different time demonstrates mode conversion occurs when the surface wave propagated through the crack. The simulation results show the relative error of determining crack depth by the proposed method is less than 6.5%.

To simplify the structure and improve the sensitivity of the optical fiber Fabry-Perot (FP) magnetic field sensor (MFS), an optical fiber extrinsic FP interferometric MFS based on Terfenol-D magnetostrictive material (TDMM) was designed and fabricated. Insert a TDMM with the diameter of 8mm and a length of 30mm into the cavity, and vapor-deposit 20% aluminum film on one end surface, and vapor-deposit 60% transmission inducing film on the fiber end surface. The two film layers are used as a FP with the cavity length of 1.7mm. The relationship between spectral drift and magnetic field sensing was theoretically analyzed. When the magnetic field intensity is 10~30mT, the sensitivity of magnetic field sensing is 0.081nm/mT and the interference signal extinction ratio is as high as 9dB at room temperature. The sensor is featured as the interferometric signal demodulation is simple and potentially can be used in engineering field.

Deep learning has excelled in feature representation and pattern classification in recent years, providing a new direction for the further development of voiceprint recognition technology. The current vocal recognition technology has been developed and widely used for a long time. Based on the analysis of the basic principles of vocal recognition, the current research status of vocal recognition technology is summarized, the current vocal recognition technology based on deep learning is analyzed, and finally the application characteristics and advantages of deep learning in speaker recognition are summarized, the problems and challenges faced by the current vocal recognition research are further analyzed, and the deep We further analyze the problems and challenges faced by the current voice recognition research, and look forward to the prospects of voice recognition research under the framework of deep learning.

Ge-Sb-Se-Te (GSST) based phase change material is an important alternative material for waveguide optical switch. We demonstrate a symmetric multiplexing 2×2 switch based on Ge-Sb-Se-Te and analyze the optical switch transmission and electro-thermal consumption characteristics at conventional wavelength from 1530nm to 1565nm. The maximum extinction ratio of 20.30dB and the insertion losses of 0.92dB are obtained for the amorphous state of the phase change material, 18.40dB and 1.59dB respectively for the crystallized state. The amorphous transition time and consumption energy are 5.2μs and 159.1nJ, and the corresponding values are 1.58μs and 72.1nJ for the crystallized transition. The methods of FDTD solutions and COMSOL Multiphysics are used to build the structure model and simulate the switching transition properties.

After an uncooled infrared camera working for some time, the non-uniformity caused by detectors, optical lens and camera shells has a serious impact on the imaging. To improve performance, a single frame and multi-frame non-uniformity correction algorithm method is proposed. By the translational motion of scene, we first preprocess the image by using the low-frequency non-uniformity noise correction method in a single frame, then inter-frame images are registered by SIFT algorithm, and we calculate the compensation coefficient by using scene-based coefficient correction algorithm, and finally reduce the non-uniformity of the image. Experimental results show that the algorithm can reduce the roughness of images and effectively overcome ghost.

Remote sensing image segmentation has always been an important research direction in the field of remote sensing image processing, and it is a key step in the further understanding and analysis of remote sensing images. Image semantic segmentation is the process of classifying each pixel to form several sub-regions with respective characteristics, and extracting the objects of interest among them. However, due to the complex boundary and scale difference of the remote sensing image, the traditional algorithm can not meet the actual needs well, resulting in low segmentation accuracy. In order to further improve the accuracy of remote sensing image segmentation, this paper combines deep convolutional neural network with remote sensing image, based on the U-Net, firstly compares the model's segmentation accuracy under different learning strategies, and introduces a new learning strategy to improve the learning effect of the model; secondly, in the loss function part of the model, a new compound loss function is proposed to speed up the convergence of the network and improve the segmentation accuracy. Based on full experimental research on the WHDLD remote sensing image dataset, the results show that the improved method has 1.5% accuracy improvement compare to the U-Net.

This paper proposes an SDN address hopping algorithm based on flow count synchronization to defend against DDoS attacks. The controller and the client use the DH algorithm to negotiate to generate a hopping address pool, and use RSA to verify the legitimacy of the negotiating parties, ensuring the security of the hopping address pool, and solving the problem that the hopping pattern is easily intercepted and cracked. The flow count is used as an address hopping trigger condition, and the address hopping is selected according to the number of data packets sent in the network, which can quickly respond to large-flow DDoS attacks. Experiments have verified the effectiveness of its defense against DDoS attacks.

With the wide application of 3D face reconstruction technology in the fields of face recognition, security early warning, and target tracking in recent years, the research of 3D face reconstruction methods has become a hot spot in the field of computer vision. Although the 3D reconstruction technology has certain theoretical support and wide application basis, how to ensure the authenticity and accuracy in the process of restoring the face is still a difficult research point for a long time. This paper focuses on the 3D face reconstruction technology, mainly focusing on the 3D face reconstruction algorithm based on the single face image as input, and conducts related research. They mainly include model-based 3D reconstruction and end-to-end reconstruction algorithms. Compared with the traditional face reconstruction algorithm based on multiple images, the use of single 2D image for 3D reconstruction has more prominent advantages in terms of economy and time cost. It makes mobile 3D face offline reconstruction possible. It will bring more real and objective application prospects to the future public safety guarantee, and make the construction of smart cities possible.

During the irradiation of laryngeal cancer, the motion caused by swallowing of patients leads to the nidus deviating from the target area, which results the decrease of therapeutic effect and the waste of dose. Therefore, it is necessary to detect the displacement, time and number of swallowing during radiotherapy. In this paper, a non-contact and non-invasive swallowing detection method based on binocular vision is proposed, which detects swallowing by capturing the motion of thyroid cartilage. In our method, first a marker is used to highlight the characteristics of thyroid cartilage; Second, an improved matching method was used to match the feature points; Then the three-dimensional shape of the thyroid cartilage was established for the identification of laryngeal prominence and extraction of motion signals. Finally, the signals were processed to obtain the number, displacement and time of swallowing. Experiments were carried out on healthy people. The results show that our method is feasible for detecting the number and time of swallowing movement, and can measure the displacement in three dimensions.

At present, with the construction and advancement of domestic basic power facilities, the Internet of Things technology is slowly being effectively integrated with the traditional power grid. The effective integration of heterogeneous and multi-parameters generated by multi-source power sensing terminals is the realization of current power The key technology of the Internet of Things. At present, the integration of domestic power Internet of Things parameters is mainly based on the integration of homogeneous multi-parameters and main decision-making parameters. There are large technical gaps in the integration of heterogeneous and multi-dimensional multi-source parameters, which cannot meet the current power Internet of Things. The need for deep integration of multiple heterogeneous terminal parameters. This article analyzes the current status of multi-source power sensing terminal parameter fusion, analyzes the current power feature fusion method, fusion framework and fusion business scenarios, and finally verifies the effect of feature fusion through related scenarios.

We innovatively apply three different methods of encoding signal as 2-D plots to radar emitter signal recognition: Recurrence Plots (RP), Gramian Angular Field (GAF) and Markov Transition Field (MTF), thus the radar emitter signal recognition problem is converted into image processing problem. We build a 2-stage convolutional neural network (CNN) model to make use of its mature technology in the field of computer vision and image processing for signal recognition. These pipeline offers the following advantages: i) Encoding signal methods enable us to visualize certain aspects of the radar emitter signals through 2D images, and ii) CNN can automatically learn different levels of radar signal features and achieve a high recognition rate. In order to get a best encoding method, we compare the recognition accuracy of the three methods mentioned above, and combine the images encoded by the first two methods (RP and GAF) with high recognition accuracy into 2-channel RP-GAF images and get a better recognition rate. At last, we analyze the reasons why three encoding methods achieve different recognition rates.

Aiming at the mathematical model of ship with internal parameter uncertainty and external disturbance, in order to solve the problem of insufficient observation ability of conventional ADRC when external disturbance changes greatly, this paper combined neural network and sliding mode designing to improve and optimize the ADRC. An extended state observer is designed to estimate the total disturbance based on the nomoto model, using the method of nonlinear sliding mode design of ADRC error feedback control law, which makes the physical meaning of the parameters of the apparent ease of setting.

Laser-induced acoustic (LIA) technology is of significance for the communication from the air platform to the underwater platform. This paper describes a method of nonlinear LIA in the liquid based on the mechanism of optical breakdown. Acoustic signals with high energy were generated by using a Nd: YAG pulsed laser with 10 ns width. A PZT hydrophone was used to receive the LIA signals for cases of the linear and nonlinear regime in the water. Meanwhile, simulation of LIA in a distance of 400 m between laser spot on the water surface and the hydrophone was performed by the COMSOL Multiphysic software. The simulated results show the communicated ability of LIA technology in the long distance.

The uniformity and moisture ratio of drug mixing are the most key attributes in the production process. The whole process detection of powder mixing equipment is particularly important. The visual detection system of powder mixing equipment based on PLC is studied. In the hardware design, the transmission efficiency of detection video is controlled by multi resistance acquisition circuit. In the software design, the visual model is built based on PLC, the powder mixing equipment is detected under the data exchange, and the visual detection system is designed. Experimental results: under the test environment of completing the detection of 3 million drugs every day, the average number of false kicks in this system is 20. The amount of false kicks has been effectively controlled and can meet the actual production needs.

A Convolutional Neural Network (CNN) for theoretical station area line loss is proposed in this paper. Considering that CNN has strong nonlinear fitting ability, it is often used to predict the station area line loss. We analyze case, and select appropriate number of input features to verify proposed method’s availability. Meanwhile, the station area line loss is calculated under the most appropriate number of feature inputs. The results show that the station area classification and key factors are identified as the subsequent station area loss calculation model, which optimizes the input variables and improves efficiency and accuracy for station area line loss calculation.

Artificial intelligence is playing an increasingly important role in today's society. The main topic of this article is finding the effects that machine learning has on improving artificial intelligence in specific ASR(automatic speech recognition) and path planning fields. By analyzing various examples and case studies, the author found that the use of artificial intelligence learning methods can make the program calculation faster and obtain better results in a relatively short time. And this improvement can be considerable when taking large number of calculations and if autimatic driving could be used in the future, this would save some time on the road or save passengers’ lives because of the decreased reaction time.

In order to realize the comprehensive visual inspection of planetary rovers, a highly integrated photoelectric payload including separation mechanism and WiFi camera was proposed. The WiFi camera can be detached from the rover through the separation mechanism. Power supply module was designed to achieve energy self-sufficiency of the camera. WiFi communication module was put forward to implement information transmission between the camera and the planetary rover. Several tests were carried out and showed that the photoelectric payload design was reliable and feasible.

In order to satisfy the requirement of communication signal coverage and reduce the investment, the optimal grazing angle of traffic arteries communication signal coverage based on power lobe rotation is obtained. The main lobe power angle, the power lobe rotation function and the critical grazing angle are determined by calculating the vertical distance spatial propagation loss and vehicle body penetration loss; Then the critical grazing angle of vertical distance communication signal coverage is calculated. The minimum power angle and optimal grazing angle of the main lobe satisfying the coverage index are determined. The power angle lobe strength and the corresponding different vehicle body losses of different grazing angles are comprehensively considered to satisfy the coverage level index and ensure the maximum effective coverage distance of a single sector, so as to maximize customer satisfaction, reduce the number of construction base stations and maximize the return on investment.

Whispery voice is a type of voice quality in which the vocal folds do not vibrate and the airflow through the resonating cavity is modulated. Previous studies have shown that there is no fundamental frequency but has formant envelope in whispery voice. This is why the listener can recognize the voice quality. In this study, acoustic and spectral measures are extracted to investigate the formant pattern and voice quality of whispery voice. The results show that the differences in harmonic amplitudes (H4-H2K, H2K-H5K) are more effective in distinguishing whispered speech from modal voice than the differences in harmonic amplitudes at low frequencies (H1-H2, H2-H4). In addition, the values of Harmonic-to-Noise Ratio (HNR) and Cepstral Peak Prominence (CPP) of whispered voice were significantly lower than that of modal phonation, and the acoustic energy of whispery voice was also significantly reduced. As to formant frequency, the lower formant frequency of whispered vowels became higher compared to those of modal phonation. These findings can not only reveal the acoustic characteristics of whispery voice, but also provide theoretical foundation for whisper automatic recognition.

This paper studied the style and characteristics of traditional woodblock New Year pictures, explored the potential of recreation and development of traditional Chinese zodiac signs, applied computer graphics and image processing software technology, comprehensively extracted, reconstructed and redesigned the graphics and color elements of the Chinese zodiac and New Year pictures, and further researched and created a unique design. The twelve zodiac graphics with Chinese tradition and national characteristics realize the combination of tradition and modern artistic aesthetics. Let more modern people re-recognize the Chinese zodiac and traditional New Year pictures, and call on people in modern society to pay more attention to and protect the traditional Chinese zodiac culture.

During the 23rd collective learning of the Central Political Bureau, General Secretary Jinping Xi commented: “There are cultural relics in museums, heritage on the vast land, and written words in ancient books. Let’s make them alive”. [1] Following the national policy that promotes the digital protection and scientific application of cultural heritage, this paper investigates the methods used to digitize fossils, which involves the integration of palaeontology and arts. We aim to increase the creativity and vitality of the displays in museums, facilitating scientific communications. We also discuss how interdisciplinary approaches would benefit the transmission of science. Through collaboration between museums and institutes that transmit and research science respectively, we hope to achieve collaborative developments including creative popular science, art popular science, conversation popular science and transmitting popular science. This paper uses Jehol Biota from Western Liaoning, China as a case study to illustrate the proposed methods. It is the birthplace of the internationally renowned Sinosauropteryx, as well as numerous feathered dinosaurs, which have been crucial to understanding the origin of birds. [2] We explore the methods and pathways of transforming museum resources into creative products. With digital reconstruction, data could be used for research, as well as scientific popularization, using the art-transformed resources for scientific communications.

In order to make the ultraviolet communication network meet the requirements of connectivity, energy saving, less interference and fault tolerance, topology control technology is a very effective method to adjust the transmission power of nodes. In reality, the application scenarios of wireless ultraviolet light communication are complex and changeable, and the optimization goals of the network may also change accordingly. A fault-tolerant three-dimensional topology control algorithm that is closer to the real environment is proposed in this paper. Under the premise of topology fault tolerance, the algorithm can optimize the interference to the minimum when the interference has a large impact on the network and can better consider energy saving when the interference impact is small.

Due to the flexibility and maneuverability of unmanned aerial vehicle (UAV), it is convenient and effective to deploy multiple UAVs as aerial base stations to provide wireless services to ground users(GUs) at some emergencies (such as disaster relief and overcrowding). However, because of the limited capacity and electricity of UAV mounted base station (UAV-BS), it is a challenging problem that how to deploy UAV-BS for on-demand coverage, especially the area to be covered is large which need multiple UAV-BSs to cooperate to provide wireless connectivity. The different characteristics of terrestrial channel of ground BS and air-to-ground channel of UAV-BS makes it a new deployment problem. In order to make full use of the energy of the UAV-BS, this paper proposed a Cyclic K-means coverage algorithm and the deployment of the UAV base station in three-dimensional space was solved while considering the quality of service and the load capacity of the UAV base station. Simulations are conducted to evaluate the performance of the proposed method for different distributions of the GUs.

Synthesized RF system is the development direction of complex electronic system platform in the future, and the implementation of synthesized RF system is the key technology. This paper focuses on the multi-beam forming technology of UWB synthesized RF system. The cost of hardware implementation based on the traditional TTD (truetime- delayer) beam-forming technology is high and the scanning angle range is limited. In this paper, we firstly address a multi beam-forming approach based on true time delayers (TTD). Then we build a lens-based array system which can form high gain beams with a small number of antenna elements and also establish the mathematical model of the lens. Lastly, we proposed a radiation pattern synthesis method to scan continuously and radiate multi-beams of different function signals simultaneously. Simulation results verify the effectiveness of the proposed implementation and scanning method.

A wideband Magneto-Electric dipole antenna fed by grounded coplanar waveguide (GCPW) is presented in this paper. The antenna is a two-layer structure, which contains a I-shaped slot on the top surface of the SIW as a magneto dipole, and several rectangle patches connected to the top surface of SIW as electric dipole. A wideband 2x2 ME dipole antenna array also is shown in this article. The feeding network is consist of a GCPW to SIW transition and a T-shaped divider with 180° selfcompensating phase shifter. In addition, the parasitic patch on the middle of top layer and several holes in the feeding network are also adopted to improve the wideband. The proposed antenna element in this paper achieves an impedance bandwidth of 36.2% (28.9-41.7GHz) and average gain of 7dB. The measured impedance bandwidth of the antenna array is16.4%(29.1- 34.3GHz) with peak gain of 13.2dB.

In response to the problem of the single event upset(SEU) that occurred during the year of the calibration satellite’s orbit, the satellite’s accumulated SEU data from December 2018 to January 2021 was selected as a sample, based on the analysis of satellite orbit data changes. Detailed statistics on the monthly, quarterly, and semiannual data statistics of the occurrence time of the SEU and the space-time distribution characteristics of the sub-satellite point, and finally determined the relationship between the SEU of the calibration satellite and the distribution of the earth's magnetic field, the F10.7 curve, neutron monitoring data and other factors characteristic. The results show that the probability of the sub-satellite point of the SEU of the calibration satellite occurring in the anomalous region of the South Atlantic can reach 49.1%, the probability of occurring in the high latitude regions of the south and north levels is 33.14%, and the other regions are less than 17.71%, and the SEU of the calibration satellite. From this, it can be inferred that the corresponding aging of the reinforcement devices which are related to single particle radiation on the satellite will cause the probability of SEU increase.It is predicted that SEU will become an upward trend in the following orbit operation. Further monitoring is needed to ensure that the standard satellite function is normal.

In order to achieve rapid and accurate detection of infrared power equipment images, this paper proposes a lightweight target detection model for real-time detection of infrared power equipment images based on the target detection network YOLOv5, which enhances the attention module and improves features by incorporating local spatial embeddings Methods such as extracting the network can improve the detection accuracy and speed of the model. Experiments show that compared with YOLOv5, the method in this paper has a higher detection accuracy while the detection speed is similar, which provides a new idea for the intelligent detection of infrared images of power equipment.

At present, most of the routing process of communication network is based on the mathematical model. This method can easily lead to long transmission delay of computer communication network. For this reason, the optimization method of computer communication network routing based on tabu search is studied. By analyzing the existing routing problems, a mathematical model is built based on Tabu search, and a genetic algorithm is used to implement the routing of communication network. The experimental results show that the transmission delay of the computer communication network using the research method is 15.09m/s and that of the comparison method is 19.89m/s. This verifies that the research method can effectively reduce the transmission delay of the communication network and improve the performance of the computer communication network.

The shallow water areas, such as near shore, near reef, shallow sea and wetland, are blind areas for detection, and are also important research fields of LiDAR bathymetry technology. In this paper, we used the self-developed LiDAR bathymetry experimental system to obtain echo waveforms in the air and water tank in the laboratory. A deep learning method of one-dimensional convolutional neural network was proposed to directly invert the depth of water based on these original experimental echo data. The results denote that the deep learning method is feasible for the processing of large amount of LiDAR bathymetry echo waveforms. This method can be further used for echo waveforms processing outdoor.

To solve the problem of low recognition efficiency under the low SNR of radar emitter signal recognition, CNN and LSTM are used to realize the recognition of signals in different intra-pulse modulation modes. Firstly, this paper achieves the local characteristics of the signals with CNN. Then capture the global characteristics with LSTM. Finally, construct the logical regression classification to complete the classification and recognition task. The simulation results show that when the SNR is -6dB, the overall recognition accuracy can reach 98%, and when the SNR is greater than -2dB, the accuracy rate is up to 100%. To verify the effect of different CNN layers and LSTM layers, comparative experiments are carried out. The results show that the appropriate increase of convolution layers is beneficial to improve the accuracy, while the lack of LSTM was not conducive to classification and recognition.

In order to improve the collision avoidance capability of unmanned aerial vehicle (UAV) formation, a new kind of UAV formation control algorithm was developed based on combining the consensus control and the artificial potential field (APF). The formation controller consisted of three sub-controllers. The first was a distributed consensus controller that controlled UAVs to maintain the desired formation shape. The second was an APF controller that avoided collisions among UAVs. The third was an APF controller that avoided collisions between UAVs and obstacles in the task space. Numerical simulation was performed to validate the effectiveness of the designed UAV formation controller. The simulation results show that 5 UAVs can safely avoid collisions with each other and collisions with the obstacles during their formation flight. Moreover, they can recover desired distance and desired shape after they avoid the obstacles.

IEEE 802.1AS is a time synchronization standard that can be applied to distributed networks and real-time control systems. It exchanges time synchronization messages based on Ethernet to achieve high-precision time synchronization. The IEEE 802.1AS protocol stipulates that when the network structure does not exceed 7 hops, the time synchronization accuracy is less than 1μs, but the different positions for obtaining the time stamp will lead to different accuracy of the time synchronization system. Based on the analysis of the IEEE 802.1AS protocol and the existing time synchronization system, this paper proposes a time synchronization system implemented by hardware logic, and obtains the time stamp at the MAC layer to achieve higher time synchronization accuracy. Two ZCU102 FPGAs are designed as master and slave nodes respectively. Through behavior simulation verification, the synchronization accuracy between master and slave nodes can achieve an average error of 6ns~50ns, which meets the demand for high-precision time synchronization in complex environments such as aerospace.

In the complex electromagnetic environment of the battlefield, it is particularly important to ensure the communication network of the UAV cluster and the safety of the UAV. This paper proposes an algorithm for avoidance wireless UAVs in the UAV cluster. First, the enhanced vector field histogram algorithm is improved by combining the characteristics of airborne wireless UV coverage and the characteristics of UAV motion. Secondly, using a traceless transformation to improve the one-step prediction in the Gaussian mixture probability hypothesis density filtering algorithm. Simulation results show that compared with the A * -based enhanced vector field histogram algorithm, the overall trajectory of the algorithm in this paper is smoother and there is no obvious jitter when avoiding. Prove the feasibility of this algorithm.

In this paper, the problem of large positioning error of infrared coordinated targets for two aircraft formations is studied. First, based on the combination of spherical coordinates and rectangular coordinates, a mathematical model of infrared sensor cross-co-location positioning is established. Then, the kernel ridge regression algorithm is used to improve the singularity of the positioning matrix, and two sets of data with higher precision are selected to calculate the positioning error covariance matrix. Simulation analysis shows that the algorithm can significantly improve the positioning accuracy in the entire detection area.

The purpose of this study is to develop an intelligent automatic scoring system of Tai Chi 12 GONG FA based on machine learning to assist the teaching, training and promotion of 12 GONG FA. On the basis of obtaining the action videos of the practitioners with different exercise levels, the two-dimensional human skeleton key point recognition technology was used to extract the different behavior patterns of fine and bad actions by analyzing the complex periodic changes of the spatial distance of different key points in the time domain; And the artificial neural network was used to reduce the dimension of features predict whether the actions conform to the norms. The system only uses monocular RGB video as input, and by the proposed algorithm of recognizing fine action and bad action, 12 kinds of action can be well identified. The intelligent scoring system based on key point detection achieves similar results with the experts' score.

At present, video communication is mostly two-dimensional communication, and holographic immersive communication cannot be realized. With the advent of the 5G era, relying on the high-speed and low-latency characteristics of 5G networks, holographic communication that integrates human-computer interaction technology, artificial intelligence, sensor technology, holographic projection display technology, etc., can completely break the boundaries between local and remote spaces, and provide users with an immersive communication experience. In this paper, we combine holographic imaging technology to introduce the definition of holographic communication, and summarize the exploration results of major research institutions in this field in recent years. Finally, the system framework and key technologies of holographic communication are analyzed. Through the investigation of the difficulties in holographic communication technology, we have summarized the future technical research directions.

Recently, many artificial bee colony-based localization algorithms have been used to obtain more accurate location information. However, the situation of co-linear anchor nodes and measurement errors lead to the problem of flip ambiguities. In the current researches on the localization algorithm of artificial bee colony, the lack of detection and correction mechanism for flipped nodes causes large errors. In this paper, we propose an improved artificial bee colony localization algorithm. We firstly qualify the initial search area based on the measured distance. Then, we introduce a robustness criterion to evaluate the localization nodes, which will avoid using the possible flipped nodes as assistant anchor nodes and thus improve the localization accuracy. Simulation results show that our algorithm can effectively avoid the flip ambiguities and achieve higher accuracy than existing ABC-based and other metaheuristic-based localization algorithms.

In order to realize the automatic planning and automatic control of path tracking error of robot with dual-mode rigid body unit hybrid control, an automatic planning method for path tracking error of robot with dual-mode rigid body unit based on external parameter expression parameter calibration and parameter fusion is proposed. The motion parameter distribution model of the hybrid control robot with dual-mode transmission rigid body unit is constructed, the motion information of the hybrid control robot is sampled by adopting the spatial attitude parameter acquisition method of the dual-mode transmission rigid body unit robot, the adaptive area parameter control of the hybrid control robot with dualmode transmission rigid body unit is carried out by combining RBF neural network and external parameter expression calibration algorithm, the kinematics analytical model of the dual-mode rigid-body unit hybrid control robot is established by using the method of 3D vision distributed fusion. The structural parameters of the dual-mode rigid-body unit hybrid control robot are identified by autonomous fusion control of structural parameters and motion parameters, and improved external parameter expression calibration algorithm and attitude parameter fusion method, thus realizing path tracking control and error adjustment of the dual-mode rigid-body unit hybrid control robot and realizing automatic path planning. The test results show that the automatic path tracking error planning of the robot with dual-mode transmission rigid-body unit hybrid control has good adaptability and low proofreading error.

With the rapid economic development of China, the demand for electricity from industry and residents continues to increase. The state vigorously develops basic work related to power operation and maintenance. But the question is how to effectively troubleshoot faults in the face of China's complex geographic environment. It is the key to improving the efficiency of power transmission. To improve the overall effect and level of the power system, it is necessary to pay attention to the development and maintenance of the power operation and maintenance system, and to continuously optimize and improve the communication transmission technology and positioning technology, in which the intelligent operation and maintenance system of the transmission line play an important role. This paper studies the communication and fault location technologies of the intelligent operation and maintenance system in the transmission line, analyzes the current status and existing problems of the current transmission line operation and maintenance system, and further points out that it is to meet the needs of smart grid transmission line operation and maintenance.

In the task scheduling problem of phased array radar, the scheduling algorithm based on time pointer is widely used because of its advantages such as high scheduling success rate and fine timeliness and so on. However, when the number of targets is small, the time shifting rate of this algorithm is too high to affect the tracking accuracy. For this question, a task scheduling algorithm based on backtracking strategy is proposed in this paper. Firstly, the time pointer is set up and selected the tasks which can be executable at the moment. When there is any tasks are selected, radar “backtracking” some tasks which have been scheduled and reschedule them. Finally, repeat the above steps until the end. Simulation results show that the scheduling algorithm based on backtracking strategy can reduce the time shifting rate significantly and without affecting the scheduling success rate. The effectiveness of the proposed algorithm is verified.

Based on the purpose of solving practical problems, this paper designs and explores a new type of multimedia digital platform. This kind of multimedia digital platform based on ergonomic research, composed of electronic products and the desk frame itself. And combined with the concept of humanized design and expression, so its can improve the user experience, effectively carry out multimedia teaching.

Since 2017, the rural revitalization strategy has been one of China's seven significant strategies for comprehensively building a modern and powerful socialist country. It has brought new development ideas and a powerful force for the Chengdu area to achieve the five goals of reform and development in rural areas represented by Linpan in Sichuan Policy support. This Project takes the Yongxing Linpan development project as an example, and the Project aims to what application on the Linpan settlements revitalization by using unmanned aerial vehicles (UAV) looks like, in terms of the digital survey, the building information modeling (BIM) construction, site analysis, and planning effects rendering. Why do people use UAVs as digital survey vehicles? How could people create a landscape information model by combining point cloud data and BIM? Do point cloud data caught by UAVs enhance the effects of the planning? These questions go beyond the mere forms of the digital record and object visualization to provide a unique insight into the function of UAVs in Linpan development projects.

In order to improve the utilization efficiency of cultural buildings, architects often design them as multi-purpose venues. This paper introduces the design of a multi-purpose concert hall built in Fuzhou New District. In order to meet multipurpose performances, including large-scale symphony and organ music, small-scale orchestral and solo performance, adjustable acoustic curtains have been implemented into this hall for tuning its acoustical response. The results of simulation and measurement show that this hall has adjustable reverberation in suitable range and is respond to the set design demands.

Stretchable supercapacitor with two-dimensional (2D) stretchability had high received attention for its practical applications such as wearable electronics and electronic skins. However, mechanical strength of devices and stretchable stability during repetitive movement have been seriously affected the device performance, which is all caused by the device structural design. Herein, we report a simple laser holography technology for periodic elastic substrate with 500-800μm microgrooved structures, and laser holography technology not only structured the elastic substrate but also could provide synchronous photoreduction of grapheme oxides (GO) for graphene electrodes. When the scan speed of 50mV/s, the tensile of devices is 54%, the area specific capacitance is 0.96mF/cm2, which shows less significant changes to area specific capacitance of 0.79mF/cm2 (0% tensile ratio). This work may hold great promise for production of stretchable grapheme supercapacitor that can be used in wearable electronic devices.

In recent decades, the power conversion efficiency (PCE) of perovskite solar cells (PSCs) has risen from 3.8% to 25.5%, implies its potential commercialization. However, the stability issue is still a limitation for its actual application. For efficient and stable PSCs, surface passivation of metal halide perovskite thin films has been demonstrated to be an effective way. Herein, we illustrated that by using 4-fluorophen-ethyl ammonium iodide as a fluorinated aromatic organic spacer cation (F-PEAI) for the surface defect passivation to enhance the efficiency and stability of PSCs devices. F-PEAI could effectivity decrease the defect states and suppress non-radiative recombination through the ionic interaction, leading to superior performance of the PSCs devices. The PCE was up to19.48 % after being treated with F-PEAI. The F-PEAI treated PSCs also had better environmental stability, maintained about 80% of the original PCE after storage in ambient atmosphere for 500 hours.

In order to strengthen the security of smart lock, a design scheme of plug-in key smart lock with unique identity authentication ID is proposed. Main control chip uses Bluetooth chip NRF52832, to control two encryption chips ATSHA204A. The first encryption chip is welded on the smart key and the second encryption chip is welded on the smart lock cylinder. SHA256 algorithm is used for two-way verification, and through the combination of mechanical verification and password verification, the lock can be unlocked or configured only if the key and password are paired successfully, so as to realize double guarantee and high security performance.

The vibration error mathematical model of all fiber optical current sensor (FOCS) is established. The propagation of vibration in optical fiber waveguide is analyzed theoretically, and the mechanism of linear birefringence caused by vibration is explained in essence: the stress changes in optical fiber caused by vibration propagation to optical fiber, and the linear birefringence is caused by photoelastic effect. In addition, the wavelength fluctuation caused by vibration wave propagation will cause the change of Verdet constant of the sensing fiber, which will also affect the output accuracy of FOCS. COMSOL finite element method is used to realize the coupling analysis of light field and magnetic field, sound field and stress field. The simulation results show that: the linear birefringence effect will reduce the rotation angle of the polarization plane of the incident light, and the vibration propagates to the sensing fiber, causing uneven stress distribution in the cross section of the fiber. The stress difference on the cross section of the fiber is positively correlated with the vibration frequency and amplitude. Combined with Verdet constant, the comprehensive error of FOCS under the influence of vibration is obtained, and the optimization suggestions are given.

The cathode for molten salt electrolysis is a pure tungsten rod formed by powder sintering, which has a high melting point and poor welding performance. The traditional welding process cannot satisfy the surface cladding of the tungsten cathode. This experiment uses advanced ultra-high-speed laser cladding technology, combined with computer stress and temperature field simulation, using optical microscope (OM) and scanning electron microscope (SEM) for bonding surface observation, using energy spectrum (EDS) for element diffusion analysis, and using display Microhardness tester analyzes the performance analysis of the bonding layer. The results show that the use of ultra-high-speed laser cladding technology can realize the corrosion-resistant layer cladding on the tungsten substrate. The use of powder with a thermal expansion ratio equivalent to the substrate and the preheating process before the tungsten substrate cladding can effectively avoid internal stress. The phenomenon of cracking; through observation of the bonding interface, it can be obtained that the cladding layer is well bonded to the substrate, and the cladding bonding layer formed has no cracks at the interface.

In the process of coal mining, it is easy to cause geological disasters such as ground collapse, which is particularly important for the evaluation and prediction of the risk of ground collapse in the mining area in order to reduce the losses caused by ground collapse. Ground collapse is affected by geology, hydrology, weather and other factors, the paper puts forward the risk evaluation model of ground collapse in coal mining area based on LTMS, takes 10 characteristics of pretreated mining area collapse as LSTM model input, according to the input data quantity, selects the number of samples per batch after experiments is 8, and selects Adam as the optimization function by SGD, RMSprop and Adam, and the number of hidden layer nodes is set to 30. The LSTM model achieves optimal performance at dropout value 0.5. The output of the model is divided into four levels of risk evaluation of the mining area, so as to construct the risk evaluation model of ground collapse in the mining area based on LSTM. It provides a new feasible way to evaluate the risk of mining goaf.

The popularity of intelligent devices has driven the application and development of Internet of Things (IoT). Data generated by sensor networks are more inclined to be transmitted and processed in the form of streaming data. The appearance of edge computing can effectively relieve the pressure of network bandwidth. However, at the edge of the network, the traffic is often uncertain, which leads to the fluctuation of network bandwidth and load. Therefore, it is of great significance to predict the future traffic according to the previous data transmission volume and current data volume of each node of the network. In order to improve the efficiency of edge unloading and the resource utilization of edge computing server, a method based on long short-term memory (LSTM) model and grey model is proposed to predict the traffic over a period of time, and the device and equipment corresponding to the method are given. The simulation results show that the method can effectively solve the data storage and calculation of IoT terminals, and improve the accuracy of IoT edge traffic prediction.

Choosing the more humorous edited headline is a subfield of humor detection and generation tasks. This paper tries to deal with the second subtask of the SemEval-2020 shared task, “Assessing Humor in Edited News Headlines”. It aims to determine how machines can understand humor generated by an atomic edit to the original headline and automatically pick up the funnier version among two different edits. Given that both substitute words on the same original text are scored using crowdsourcing, we attempt not only classification but also the regression model on this special task. As for the training process, we first consider using two different embedding approaches, including GloVe and BERT, then further use different forms of neural network such as a fully connected layer, BiLSTM, and GRU. According to the experimental results, our BERT-based model gets a 64% accuracy performance, ranking second in the competition over 50 teams. Furthermore, by comparing the result and performance of the above models, we pick up some classic wrongly predicted samples and analyze the potential reasons for future study. The experimental results illustrate that mainly the revised sentence accounts for edit humor, whereas the original sentence does not have any effect. Besides, the combination of the revised and original sentence as input receives the best output, which shows that edit humor is probably produced from the edited sentence and the difference before and after modification.

This paper presents an improved ultrashort wave single-channel Doppler direction finding system. The circuit method of comparator + mid-band pass filter is proposed and analyzed to eliminate the influence of incoming wave amplitude modulation information on direction finding. An electronic switch switching frequency adaptive algorithm is proposed to eliminate the influence of the arrival frequency (phase) modulation information on the direction finding, that is, the method of using software to control the electronic switch switching frequency, programmable bandpass filter center frequency and bandwidth. When the switching frequency of the electronic switch is zero, the software calculates the appropriate switching frequency of the electronic switch and the narrowband bandwidth of the programmable bandpass filter based on the analysis of the arrival wave spectrum characteristics to give the horizontal angle of the arrival direction. This article gives an algorithm for the horizontal angle of the arrival direction.

In view of the large energy loss problem of common mechanical refrigerated vehicles on the market, an improved adaptive genetic algorithm(AGA) is proposed to apply to the energy consumption optimization of the refrigeration system of electric refrigerated vehicles, and the crossover operator and mutation operator are adaptively adjusted through nonlinear adjustment. Probability enables the algorithm to generate better individuals during the evolution process to obtain the global optimal solution, which minimizes the energy consumption of the refrigeration system. This paper takes a certain type of electric refrigerated vehicle as the research object, uses TRNSYS to model the electric refrigerated vehicle compartment and its refrigeration system, simulates the annual refrigeration conditions, and uses an AGA to optimize the blower flow rate and refrigerant delivery pump flow rate. AGA can reduce the total energy consumption of refrigeration system by 9.8% and improve the COP by 4.7% compared with GA. The improved AGA enhances the local search ability of genetic algorithm (GA), improves the efficiency of the algorithm, and effectively reduces the energy consumption of refrigerated vehicle.

Vanadium pentoxide (V₂O₅) film has thermal sensitive, gas sensitive, electrochromic and thermochromic. It is a common and easy to process material. The thin film samples of vanadium oxide were prepared by sol-gel method in this paper. The flocculation precipitation in the precursor is amorphous V₂O₅, which has no effect on the film thickness after filtration. The surface defects of the film were reduced by means of lamination and spin coating. The relationship between the flatness of vanadium oxide and the improvement of electrochromic performance is analyzed. The cyclic stability of vanadium pentoxide films was improved by 10% using polymer tetrabutylammonium perchlorate as supporting electrolyte. Tetrabutylammonium tribromide only increases the cyclic stability of the films by 2.3% due to the instability of the Br^-.

In order to apply in the optical path of large incident angle, an experimental platform based on 61-element unimorph deformable mirror (DM) is fabricated to study the influence of different incident angles on the correction ability of DM. The experimental results show that the larger the incident angle is, the smaller the reconstructed Zernike amplitudes are and the larger the normalized residual error is. Furthermore, a DM driven by an elliptical PZT is designed and fabricated, and its correction ability in the high incident angle optical path is tested experimentally. The results show that the designed elliptical PZT can effectively improve the correction ability of the DM in the vertical optical path.

In recent years, with the development of the domestic financial market and the growth of investors, a large number of online automatic trading platforms have appeared, and the number of investors using automatic trading systems has increased rapidly. It can be seen from the penetration rate of foreign automated trading that China's automated trading has very broad application prospects. Starting from the author’s trading experience, the article builds an automatic trading system based on the transaction model that selects advantageous trading opportunities and increases positions. The system has the advantages of good entry position, low risk, reasonable Sharpe ratio, etc., and has good application value.

Sometimes, the high-resolution A/D converter needs more ADC cores in one A/D chip. At this time, the area of layout will become the key factor of A/D design alternative. Except layout design, it also needs some special schematic structure to decrease the layout of A/D chip. In SAR A/D convertor, the core units are SAR capacitor array and comparator. In typical comparator apply, the comparator is the two-ports comparator, and one port of compatator will connect the SAR capacitor array, another port will connect a equal capacitor array. Although this comparator has the high resolution in compare stage, it also costs more layout area. This paper will introduce a real single-port comparator unit schematic structure in SAR A/D convertor, it only need one SAR capacitor array to decraese the layout area, and it also can identifies 3uV in compare stage. At last, through simulation verification and finished circuit test, this comparator in this paper can saisfied the apply of 16-bit SAR A/D convertor.

Research and design the temperature controller of dry-type transformer, using three-way four-wire PT1000 thermal resistance as the temperature sensor, STC8A MCU as the control unit and ADC, and using the recursive average filtering algorithm to process the signal. The design uses five digital tubes to display information such as temperature. It adopts forced air cooling (AF) and is directly powered by 220V power supply. In this way, the purpose of accurately controlling the transformer to work within the normal temperature range is achieved.

Corporate failure prediction is based on the extraction of key factors and early warning based on factor indicators or characteristic variables related to corporate failure. The current representative failure prediction empirical models include Credit Scoring, DA, MDA, and so on. However, the above models have problems such as data deviation and uncertain factor contributions. It is to explore starting from the machine learning model, establishing a combined model, combining the minimum variance method, to construct the Logistic-SVM combined model. The study found that the total classification accuracy of the Logistic-SVM combined model is higher than that of the traditional single model, and the error rate is also lower.

A method for predicting abnormal behaviors of substation workers based on video scenes and using generative confrontation networks to integrate global and local information is proposed. In the substation, this method can be used to issue timely warnings to the transportation and inspection personnel that may trigger dangerous actions during the operation, so as to provide an important guarantee for the life and safety of the transportation and inspection personnel. The human behavior prediction task aims to predict future behavior video frames based on a given behavior video frame. Considering that the human behavior video contains not only relatively stable scene information, but also time-varying and complex human behavior information, this method first uses a global generation confrontation network to generate video scenes and rough human contours; then uses local generation confrontation Network to further optimize the details of human behavior in the video. Experiments show that, compared with the existing methods that only use a single model to achieve pixel-level behavior prediction, the method of combining global and local generation proposed in this paper can better capture the spatial appearance and the timing dynamics of humans in the video.

In the substation, inspectors will lead to electric shock and other safety accidents when they are too close to substation equipment. It is extremely important to monitor the safe distance between inspector and equipment. This paper designs a safe distance monitoring method based on machine vision. The method mainly applies monocular camera and machine vision technology on the safe distance detection. Firstly, the safe distance is set according to the power safety code. Then the distance function is decided by human body key points and measured distance. The human body key points are achieved using human pose detection model. Through the function fitting, distance function is set as an univariate high-order linear function. Finally, human ankle key point is chosen and inputted into the distance function to determine the distance between inspector and substation equipment. Compared with the safe distance, the system can judge the inspector is in the safe state or not. Experiments show that the method can fulfill the requirements of distance monitoring for inspectors in substation, and provide a solution for the development of intelligent substation.

In recent years, quantitative trading has rapidly become popular in the domestic investment field, and it has played an increasingly important role in the investment decision-making of private equity funds and investors. This article combines the traders’ quantitative trading experience to build a quantitative trading system that repeatedly trades at the same price to capture the trend market from the aspects of trading models and position management. The article describes the use of the quantitative trading system in combination with the parameters, and the test results show that the system has good profit potential.

In view of the complexity, high dynamic, network harshness, resource limitation, strong antagonism and high damage in the environment of edge information service nodes, and the resources and capabilities of nodes are different greatly, this paper proposes a continuous support oriented edge service random access technology. Firstly, the security authentication architecture model for edge nodes is constructed, and the cross node access authentication technology in mobile environment is adopted to realize the random access capability of mobile users between cloud edges and edge edges. Secondly, the service / data continuous guarantee mechanism for edge nodes is established. By designing the mobile edge computing architecture and combining with the data migration strategy for network transmission, the edge node's random encounter, cross node access and service / data continuity guarantee capability in mobile environment are improved. Finally, an experimental environment covering network simulation environment, cloud center environment and edge information service environment is built. The experimental results verify the effectiveness and feasibility of the edge service casual access technology for continuous support.

Nowadays, real-time video is widely used in many fields such as video calling, live video broadcasting. The delivery latency and reliability requirements for live video are stringent, and using wireless access network to provide real-time video services brings great challenges to the provision of such services. In order to solve these problems, this paper proposes a mechanism to support real-time video service on multi-interface wireless devices by utilizing two selective redundancies simultaneously. The mechanism dynamically adjusts the video encoding rate and uniformly FEC encodes, then copies some video frames selectively according to the survival time of the video frames and the impact on the quality of the video, and further arranges their transmission (or part of them) with the help of the neighbor devices. In order to achieve the highest quality of video streaming service when the maximum allowable network bandwidth is given, the mechanism proposed in this paper will dynamically optimize encoding rate, copy number per frame and FEC encoding mode. In order to evaluate the quality based on the original quality and duration of distortion effects, this paper also proposes an evaluation model. The numerical results show that, compared with the existing methods, this method can reduce the video distortion to a certain extent when the network load capacity is low compared with other methods mentioned in this paper.