This PDF file contains the front matter associated with SPIE Proceedings Volume 7489, including the Title Page, Copyright information, Table of Contents, Introduction, and the Conference Committee listing.

The soil moisture is a sensitive index which is important in the research of weather, hydrograph, zoology, etc. And its importance embodied much more in the research of distribution of the vegetation and instructing the farming and herd. This paper use the MODIS data to retrieval the Inner Mongolia region's land surface temperature and vegetation index, establishing the temperature vegetation dryness index (TVDI) model by searching and regression the characteristics sequence of the data space, and then calculated the soil moisture of Inner Mongolia region. According to the validation between the retrieval and real survey data confirmed the scientific rationality and feasibility of the model using in the region. Considering the different characteristics of the climate and geographical distribution in the Inner Mongolia region, to make the research more objective and targeted, then divided the region into four sub-districts and retrieval the soil moisture independently. Finally, obtained the soil moisture ration information based on the different characteristics of the sub-districts, and the research provided some scientific basis and efforts for the agricultural and herd work in the Inner Mongolia region.

Recently, with development of computer technology, the application field of near-infrared image processing becomes much wider. In this paper the technical characteristic and development of modern NIR imaging and NIR spectroscopy analysis were introduced. It is concluded application and studying of the NIR imaging processing technique in the agricultural engineering in recent years, base on the application principle and developing characteristic of near-infrared image. The NIR imaging would be very useful in the nondestructive external and internal quality inspecting of agricultural products. It is important to detect stored-grain insects by the application of near-infrared spectroscopy. Computer vision detection base on the NIR imaging would be help to manage food logistics. Application of NIR imaging promoted quality management of agricultural products. In the further application research fields of NIR image in the agricultural engineering, Some advices and prospect were put forward.

In order to detect the rice kernels from color rice images precisely, it is necessary to remove noises from the original images very well by effective denoising methods. This paper proposes a new algorithm of wirelike noise removal according to its characteristics in rice kernel images based on the color space transform and mathematic morphology. The color space transform is conducted and then a simple structure element is employed as a filter to remove wirelike noise. In this way, the computation complexity in noise removal is reduced a lot while keeping detailed textural information well and improving the quality of images. Experiment results demonstrate the effectiveness of the algorithm.

Automatic food classification with digital images has played an important role in modern agricultural and food engineering. For this purpose, a kind of recognition algorithm for food is presented based on their shape and texture information in this paper.. By using a combination of shape and texture feature, improved mean-shift procedure is a state-of-the-art learning algorithm for multi-classification of food. The proposed method has four steps: (1) computation of a high contrast monochrome image from an optimal linear combination of RGB components of the food colour image;(2)a morphological shape detection operation is applied to detect the actual food shape from the high contrast monochrome image,some structural elements that have special forms are utilized to eliminate noise and improve detection precision; (3)a food texture is modeled by co-occurrence matrix;(4)a feature combination method is specified by food shape and texture information synthetically, then an improved mean-shift algorithm is proposed to achieve automatic food classification and recognition. The algorithm was implemented in Matlab and tested with 180 images (512×512) taken for various food with big differences. The algorithm can be applied to recognize food categories at the speed of 1.13s per image with the approval recognition rate of 97.6%. The result shows that our algorithm fully satisfies the requests of real application.

Synthetic aperture radar (SAR) images are inherently affected by multiplicative speckle noise, which is due to the coherent nature of scattering phenomena. This paper presents a despeckling method for SAR images based on adaptive bandelet transform. Bayesian maximum a posteriori (MAP) estimation is applied to adaptive bandelet transform coefficients to achieve more satisfying results. The performances of adaptive bandelet transform and wavelet thresholding for despeckling SAR images are compared through an experiment. Experiment results clearly demonstrated the capability of the proposed scheme in SAR image speckle reduction especially for SAR images possessing detailed textures.

Road information in RS images has high application value in many fields. Much attention has been paid to road extraction in RS images in recent years, and a lot of algorithms have been developed. But most of algorithms are limited to certain type of RS images, or feature low rate of recognition, and so on. So far users still can't find a satisfactory method. In order to meet the demands on data acquisition of GIS application, we developed a new extraction algorithm on road border. In this method, all roads are classified into two categories: roads with complete information and roads with incomplete information. At beginning, binary image is achieved through simple pretreatment on the whole RS image. With the initial seed point given artificially, then the whole road borders will be searched out rapidly making use of the five-neighborhoods searching algorithm. Roads that can't be searched successfully in above steps are named as roads with incomplete information. To these roads we add certain extra processes and extract borders again.

To calibrate in vision-guide motion control system, the author provides the whole processes for this. Calibrating the vision system to a real-world unit, first of all is to Correct for Common Forms of Distortion, then to Correlate Image Units to Motion Control Units, finally to determine how large the offset angle is between the two coordinate systems. It gives every step in details, which exceeds traditional method to calibrate the vison.

The remote sensing can help monitoring the nature scene, so it is very useful for agriculture engineering. This paper presents a method to simulate the SAR image of nature scene by the ray tracing and statistical method. The scene is composed of the tree and grass which would be modeled by this two method. The result shows that this method can simulate the SAR image effectively.

The detection method of pests in stored grain is always investigated. The method based on image recognition is often discussed. With development of computer technology, information processing, pattern recognition, intelligence detection, detection method based on image recognition develops fast and becomes main direction of grain pests intelligence detection. This paper puts forward an edge detection algorithm based on grey relation analysis.The importance of image's edge detection based on grey relation analysis in pests image processing is introduced. At first the reference series and compare series are defined. Then the relevant coefficients between the reference series and compare series are calculated to every pixel. Finally, the edge detection is processed and its application in image's of pests in stored grain is discussed. The examples show that the method can detect the image's edge of pests in stored grain better.

Orchard is an important agricultural industry and typical land use type in Shandong peninsula of China. This article focused on the automatic information extraction of orchard using SPOT-5 image. After analyzing every object's spectrum, we proposed a CART model based on sub-region and hierarchy theory by exploring spectrum, texture and topography attributes. The whole area was divided into coastal plain region and hill region based on SRTM data and extracted respectively. The accuracy reached to 86.40%, which was much higher than supervised classification method.

Homogeneity has same or similar shape is so common in the abstract and in nature, and shape similarity is a very important factor for classification and object recognition. Traditionally, a multidimensional vector is treated as a point of the feature space, we calculate the distance between the points to measure the similarity, the smaller the distance, the greater the similarity. The popular similarity measures maybe the Manhattan and Euclidean distances. In this paper, we showed the Minkowski metric computed by the absolute difference of vectors, and ignored the characteristic of the differences. According our previous works, we used objects but not points to respect the vector in the feature space, then the shape similarity can be respected by the character of the differences between vectors. Based on this point, a quasimetric distance was used for similarity estimation. Experiment results on two benchmark datasets from the UCI repository showed this kind of distance can achieve higher accuracy than the classical Manhattan and Euclidean distances in similarity estimation.

The detection of faint, small and hidden targets in synthetic aperture radar (SAR) image is still an issue for automatic target recognition (ATR) system. How to effectively separate these targets from the complex background is the aim of this paper. Independent component analysis (ICA) theory can enhance SAR image targets and improve signal clutter ratio (SCR), which benefits to detect and recognize faint targets. Therefore, this paper proposes a new SAR image target detection algorithm based on ICA. In experimental process, the fast ICA (FICA) algorithm is utilized. Finally, some real SAR image data is used to test the method. The experimental results verify that the algorithm is feasible, and it can improve the SCR of SAR image and increase the detection rate for the faint small targets.

In this paper, multispectral imaging technique for plant diseases diagnosis is presented. Firstly, multispectral imaging system is designed. This system utilizes 15 narrow-band filters, a panchromatic band, a monochrome CCD camera, and standard illumination observing environment. The spectral reflectance and color of 8 Macbeth color patches are reproduced between 400nm and 700nm in the process. In addition, spectral reflectance angle and color difference is obtained through measurements and analysis of color patches using spectrometer and multispectral imaging system. The result shows that 16 narrow-bands multispectral imaging system realizes good accuracy in spectral reflectance and color reproduction. Secondly, a horticultural plant, cucumber' familiar disease are the researching objects. 210 multispectral samples are obtained by multispectral and are classified by BP artificial neural network. The classification accuracies of Sphaerotheca fuliginea, Corynespora cassiicola, Pseudoperonospora cubensis are 100%. Trichothecium roseum and Cladosporium cucumerinum are 96.67% and 90.00%. It is confirmed that the multispectral imaging system realizes good accuracy in the cucumber diseases diagnosis.

Information extraction and target recognition are key technologies for high-resolution remote sensing, as well as the foundation of carrying out high resolution remote sensing application. Buildings are the most important ground objects of urban areas. Therefore, the thematic information extraction of buildings from high resolution remote sensing data is of great significance in many fields. The extraction results have been widely used in urban planning, geographical data updates, population and socio-economic census, environmental monitoring and other fields. This paper proposes an algorithm based on morphological characteristics of connected components to segment image and extract buildings from high-resolution image, and successfully extracted the buildings information. First of all, select the 0.6 m pan sharpened band integrated with 3 multispectral bands QUICKBIRD image which imaged in May 2004 as experimental data, and preprocess with geometric correction and integration. Then, process images with closing and opening morphology filter in different scales and build mask to remove the background interference. Finally, use the method of gray-scale threshold, edge detection to segment and select different features to extract buildings respectively. The results proved that the object-oriented building extraction method based on morphology characteristics is superior to the general per-pixel or per-field extraction method. On the one hand, this method improves the extraction accuracy, on the other hand ,improves the contours of buildings.

In the medical domain, digital images are produced in ever-increasing quantities, which offer great opportunities for diagnostics, therapy and training. So how to manage these data and utilize them effectively and efficiently possess significant technical challenges. Thus, the technique of Content-based Medical Image Retrieval (CBMIR) emerges as the times require. However, current CBMIR is not sufficient to capture the semantic content of images. Accordingly, in this paper, an innovative approach for medical image knowledge representation and retrieval is proposed by focusing on the mapping modeling between visual feature and semantic concept. Firstly, the low-level fusion visual features are extracted based on statistical features. Secondly, a set of disjoint semantic tokens with appearance in medical images is selected to define a Visual and Medical Vocabulary. Thirdly, to narrow down the semantic gap and increase the retrieval efficiency, we investigate support vector machine (SVM) to associate low-level visual image features with their highlevel semantic. Experiments are conducted with a medical image DB consisting of 300 diverse medical images obtained from the Hei Longjiang Province Hospital. And the comparison of the retrieval results shows that the approach proposed in this paper is effective.

Agriculture Image denoising is one of important and fundamental technology in agriculture image processing. The adaptive wavelet shrinkage image denoising algorithm can determine an optimal threshold and neighbouring window size for every sub bands by the Stein's unbiased risk estimate (SURE). The P_M diffusivity completes denoising according to the direction and amplitude of gradient while as far as possible to keep the characteristic of image. A new algorithm based on P_M diffusion model and adaptive wavelet shrinkage is given through the different characteristic between those two different algorithms. This algorithm applies nonlinear diffusion to low frequency part of image decomposed by wavelet, and shrinks the wavelet coefficient by the adaptive wavelet shrinkage. Experimental results show that the new hybrid algorithm can significantly improve the denoising performances in Chinese apple image denoising.

Spatial outliers are the spatial objects with distinct features from their surrounding neighbors. Detection of spatial outliers helps reveal valuable information from large spatial data sets. In many real applications, spatial objects can not be simply abstracted as isolated points. They have different boundary, size, volume, and location. These spatial properties affect the impact of a spatial object on its neighbors and should be taken into consideration. In this paper, we propose two spatial outlier detection methods which integrate the impact of spatial properties to the outlierness measurement. Experimental results on a real data set demonstrate the effectiveness of the proposed algorithms.

For more spacious place where the human bodies are sparse and scattered, it is feasible and convenient to use algorithm of frames subtraction or background elimination which can extract the human bodies. But for the places where the human bodies are intensive such as on the buses, it is difficult to segment the bodies using this algorithm. Since the heads are more scattered than the bodies, an algorithm based on several regional growths and feature extraction is studied to detect the heads in this paper. If the clothes gray is similar to the head gray, morphologic operation is used to extract the head again. Also this paper shows the flow of the algorithm in detail. Many images taken by the bus camera were used for the experiment. The result shows that the bus passengers could be detected exactly which verified the effectiveness of the algorithm.

This article introduced a kind of orchard real-time plant disease image observation system that is composed of image gathering, data compression and wireless transmission module. According to image data of control center which gathered image information by SAA7111A, compressed data by IME6400, and transmitted to control center by TRF6903, the orchard manager proposes the reasonable solution about the orchard plant disease question. This system is suitable for the prevention of orchard plant disease, which characteristic is simple practical method, inexpensive cost, remote transmission, reliable data and stable system.

The paper propose a novel seal extract method through template matching based on the characteristics of the external contour of the seal image in Chinese Painting and Calligraphy. By analyzing the characteristics of the seal edge, we obtain the priori knowledge of the seal edge, and set up the outline template of the seals, then design a template matching method by computing the distance difference between the outline template and the seal image edge which can extract seal image from Chinese Painting and Calligraphy effectively. This method is proved to have higher extraction rate by experiment results than the traditional image extract methods.

Face recognition is an important technique which can be used in many applications. In recent years, face recognition has attracted large amount of research interest. Many recognition methods have been proposed, however, most of them are not able to make use of local salient features to effectively capture the face information. Recently, SIFT has been proposed for object matching in image retrieval area, and it proves to be a powerful matching tool. In this paper, we applied and studied SIFT method on face recognition, and compared it with the well known face recognition methods in literature, i.e., PCA and 2DPCA. Rigorous tests were carried out on 3 major face databases. Our results show SIFT has significant advantages over both PCA and 2DPCA in terms of recognition rate and number of training samples. This paper also points out some shortcomings of classic experiment method to recognize faces and improve them.

A derivative of Fisher's Linear Discriminant Analysis (FLDA), named Fisherapples for the recognition of apple lesions which is not sensitive to large variations in illumination is proposed in this paper. We make use of the linear projection that is orthogonal to the within-class scatter of the apple images from a high-dimensional image space to a considerably low-dimensional image space. It separates the data-cases well, projecting away variations in lighting. Our approach maximizes the ratio of between-class scatter to that of within-class scatter of apple lesions, i.e., we can get maximal between-class distances and minimal within-class distances after projection. This implies that the gap between the classes becomes bigger and ensures optimal separability in the new space. Besides, we take advantage of Principal Component Analysis (PCA) to project the set of apple images to a lower dimensional space in order to overcome the complication of the singular within-class scatter matrix. After that, the resulting within-class scatter becomes nonsingular and subsequently we can use standard FLDA to reduce the dimension further. Consequently, it is effortless for the computer to calculate the result. Experimental results demonstrate that Fisherapples performs better in apple lesion recognition than PCA.

Butterfly image retrieval is very important in the insect recognition research area but the existing butterfly retrieval technology presents poor performance. SIFT (Scale Invariant Feature Transform) features are reliable because they are insensitive to image scale, rotation, affine, distortion and change in illumination. The local and multiscale natures of the SIFT feature make it create better performance than other existing approaches do. In this paper, a new butterfly image retrieval algorithm based on SIFT feature is presented. The butterfly images in this research are transformed into a set of SIFT feature descriptors, and then the similarity of feature points is described by using Euclidean distance. Experimental results demonstrate that the method based on SIFT feature provides a new effective way for butterfly image retrieval. This proposed algorithm is invariant to the changes of butterfly image scale, rotation, and transformation. It is also robust to distortion and occlusion. Compared with the method of using gray histogram, the performance of butterfly image retrieval based on SIFT feature is improved significantly.

In this paper, a novel apple Physalospora recognition approach based on the Gabor feature-based principal component analysis (GBPCA) is proposed. In this method, the principal component analysis (PCA) is a powerful technique for finding patterns in data of high dimensionality and can reduce the high dimensionality of the data space to the low dimensionality of feature space effectively. Gabor filter is an effective tool because of its accurate time-frequency localization and robustness against variations caused by illumination and rotation. Three main steps are taken in the proposed GBPCA: Firstly, Gabor features of different scales and orientations are extracted by convoluting the Gabor filter bank and the original gray images. Then eigenvectors in the direction of the largest variance of the training vectors is computed by PCA. An eigenspace is composed of these eigenvectors. Thirdly, we project the testing images into the constructed eigenspace and the Euclidean distance and nearest neighbor classifier are adopted for classification. Therefore, the proposed method is not only insensitive to illumination and rotation, but also efficient in feature matching. Experimental results demonstrate the effectiveness of the proposed GBPCA.

In the agricultural research area, the study about butterflies is very important. However, there is hardly any content-based butterfly image retrieval system. The text-based image retrieval system is not objective enough, and could not provide the characteristics of image content. Conventionally, the RGB color histogram-based image retrieval can't provide spatial features of images, and is easily affected by the pixel distribution, which is unable to represent the comprehensive characteristics of images. In this paper, we proposed a new butterfly image retrieval algorithm based on keyblock distribution. The keyblock-based image retrieval algorithm is a generalization of the technology in computer image retrieval area which is very advanced and useful. Our proposed butterfly image keyblock distribution extraction contains three procedures: first, a codebook with specific length is estimated by employing the vector quantization technique; second, the original butterfly image is divided into non-overlapping blocks; third, each block of butterfly image is encoded with the index number of codebook. From the keyblocks, we can extract both the color distribution information and the local spatial information of butterfly image. In the performance evaluation, experimental results show that in our retrieval system, average recall (AR) and average normalized modified retrieval rank (ANMRR) achieved 0.74 and 0.3291, respectively.

This article proposes an evolutionary algorithm based segmentation algorithm for automatically grouping the pixels of an image into different homogeneous regions. In contrast to most of the existing evolutionary image segmentation techniques, we have incorporated spatial information into the membership function for clustering. The spatial function is the summation of the membership function in the neighborhood of each pixel under consideration. The two very important advantages of the new method are: 1) It does not require a priori knowledge of the number of partitions in the image and 2) It yields regions, more homogeneous than the existing methods even in presence of noise.

Remote sensing produces large volumes of data. In this paper, we provide a MDR (multi-directions-relation) algorithm for edge detection. Many methods reduce the noise before further detection. It might lose some useful information. We consider the method to detect the edge and reduce the noise synchronously, in order to keep more information in the image for detection. Afterwards, we discuss on the setting of the parameters carefully. One of the parameters implies the number of the edge points in one line. If you have some information about the edges beforehand, it will improve the accuracy through fixing it, otherwise, the algorithm adjust the value automatically. The experiments show that the algorithm is faithful to the source and is good at dealing the detail of images. That means this CP-based data mining method can reduce image data greatly.

The principle of singular value decomposition is introduced. Then the procedures of restoration, compression for pest image based on SVD are proposed. The experiments demonstrate that the SVD is one effective method in stored-gain pest image pre-processing.

Preserved Fruits are one of the famous and traditional Chinese agriculture foods. In this paper, we propose a method that utilizes color and texture features s for Preserved Fruits image classification. We use color moments and subband's statistics of wavelet decomposition as color and texture features respectively. A wide range of Preserved Fruits images are tested to evaluate the performance of the proposed method. The experimental results show that the scheme has produced promising results.

To detect pesticide residue on navel orange surface by nondestructive means, five group oranges sprayed water, fenvalerate, isocarbophos, fenpropathrin, carbendazim pesticides respectively were chosen as experimental samples. Laser imaging system was built for acquiring images of fruits. Unitary nonlinear regression function was fitted by analyzing gray histogram curves of images within 12-40 range. The coefficient or eigenvalue of functions was different about every navel orange. The threshold coefficient was confirmed by data processing, which can establish fruits surface sprayed pesticide or not. The result showed that laser imaging technique is feasible for detecting pesticide residue on navel orange surface.

In the measurement of object's projective area by the method of machine vision, image obtained is often defocused, because of the variety of object's thickness and the diversity of placement. The error of area measurement is increased because of the blur of defocused image. According to the principle of the edge width proportion to the defocusing amount, this paper proposes a method based on image processing algorithm to correct the error. First, the target edge image is yielded by edge detection; then, the revised target image is obtained through the logical operation of defocused image and edge image; finally, the correct area is calculated from the revised image. Experimental results show that this method can greatly improve the measurement precision, and has strong adaptability to different defocusing amount in the same image.

This paper presented X-ray of correction of testing image model based on digital image processing technique. Deformation algorithm and bilinear polynomial interpolation were applied to the distortion correction of testing image then make the corrected image smooth with combining median filter method. The experimental results have shown that this algorithm is simple, effective and able to show testing image clearly and accurately, laid a foundation for defect detection and identification following up.

A novel vision-based road detection method was proposed in this paper to realize visual guiding navigation for ground mobile vehicles (GMV). The original image captured by single camera was first segmented into the road region and nonroad region by using an adaptive threshold segmentation algorithm named OTSU. Subsequently, the Canny edges extracted in grey images would be filtered in the road region so that the road boundary could be recognized accurately among those disturbances caused by other edges existed in the image. In order to improve the performance of road detection, the dynamics of GMV and the Hidden Markov Model (HMM) was taken into account to associate the possible road boundary at different time step. The method proposed in this paper was robust against strong shadows, surface dilapidation and illumination variations. It has been tested on real GMV and performed well in real road environments.

Aiming at the change of battery location, environment light or camera location in Li/MnO2 automatic inspection process, a novel WT-FEBFNN (Wavelet Transform Fuzzy Ellipsoidal Basis Function Neural Network) approach to battery defect inspection is proposed. Firstly, WT is applied on original battery image, and low-frequency signal and de-noised signal is obtained, respectively, by setting different thresholds on different scale WT decomposition. Secondly, signal only containing defect (nick) is obtained by subtracting low-frequency signal from de-noised signal. Finally, model of FEBFNN is established and defect recognition is accomplished on 1000 battery images. Experiments have shown the proposed algorithm had a better robustness to the change of battery location, or environment light or camera location than multilayer perception(MLP), and shown that the reason for the high recognition accuracy in battery defect inspection is due to the information contents of the features as well as to proper classifier.

It is a challenging problem to overcome shift and rotation and nonlinearity in fingerprint images. By analyzing the shortcoming of fingerprint recognition algorithm on shift or rotation images at present, manifold learning algorithm is introduced. A fingerprint recognition algorithm has been proposed based on locally linear embedding of variable neighbourhood k (VK-LLE). Firstly, approximate geodesic distance between any two points is computed by ISOMAP ( isometric feature mapping) and then the neighborhood is determined for each point by the relationship between its local estimated geodesic distance matrix and local Euclidean distance matrix. Secondly, the dimension of fingerprint image is reduced by nonlinear dimension-reduction method. And the best projected features of original fingerprint data of large dimension are acquired. By analyzing the changes of recognition accuracy with the neighborhood and embedding dimension, the neighborhood and embedding dimension is determined at last. Finally, fingerprint recognition is accomplished by Euclidean distance Classifier. The experimental results based on standard fingerprint datasets have verified the proposed algorithm had a better robustness to those fingerprint images of shift or rotation or nonlinearity than the algorithm using LLE, thus this method has some values in practice.

The article researches the color difference classification of the fabric. At first, we detect the color difference of fabric, and select the L*a*b* color space by analyzing the character of different color space. In L*a*b* color space, the color difference are convert into geometric distance, with which we can calculate the value of color difference by CIEDE2000 color difference formula. the color difference classification is more complex, for neural network has the feature of adaptive learning and approaching arbitrary nonlinear function, we propose that neural network can be used on color difference classification, however, according to the disadvantages of tradition neural network which are slowly training and apt to trap into local minimum, highly flexible neural network was adopted. At last, we detect and classify the color difference of fabric samples and analyze the results of the detection and classification.

In this paper we introduce a color image segmentation system. In this system, we firstly use a fuzzy clustering method in a color space for color image coarse segmentation, and then merge the clusters use a region merge algorithm which based on color similarity and spatial adjacency. This method have implemented and tested on some applications. The results have showed the system research is encouraged.

Red rot disease is a common disease at the seedling stage of sugarcane. In order to identify red rot disease effectively, a segmentation algorithm for leaf images of sugarcane red rot disease using scan line filling is proposed. The proposed algorithm has six stages. During the first stage, the class of green plants is separated from the class of non-green plants using the color feature of 2G-R-B. At the second stage, connected regions of the class of green plants are labeled. At the third stage, outer contours are extracted. At the fourth stage, the regions surrounded by outer contours are filled using scan line filling. At the fifth stage, the images are colorized. At the sixth stage, red rot diseased spots are extracted using the color feature. The experimental results show that this algorithm can extract red rot diseased spots effectively, and the accurate rate of image segmentation for red rot diseases is 96%.

For a better and more effective way to detect lane and filter noise on road, this paper introduces a way on how to transform perspective when dealing with lane detection. An improved algorithm will change perspective from 3D to 2D and employ a modified Perspective Transform so that we can get a better affection of lane detection. And then, expected values are used to analyze key pixels to acquire accurate lane image point instead of using curve fitting commonly used in lane detection. Both of the creative measures mentioned above will help to acquire precise parameters and lane curves.

Vector Quantization is one of popular codebook design methods for speech recognition at present. In the process of codebook design, traditional LBG algorithm owns the advantage of fast convergence, but it is easy to get the local optimal result and be influenced by initial codebook. According to the understanding that Genetic Algorithm has the capability of getting the global optimal result, this paper proposes a hybrid clustering method GA-L based on Genetic Algorithm and LBG algorithm to improve the codebook.. Then using genetic neural networks for speech recognition. consequently search a global optimization codebook of the training vector space. The experiments show that neural network identification method based on genetic algorithm can extricate from its local maximum value and the initial restrictions, it can show superior to the standard genetic algorithm and BP neural network algorithm from various sources, and the genetic BP neural networks has a higher recognition rate and the unique application advantages than the general BP neural network in the same GA-VQ codebook, it can achieve a win-win situation in the time and efficiency.

In agricultural engineering, to ensure rational use of pesticide and improvement of crop production, computer image recognition technology is currently applied to help farmers to identify the degree of crop diseases. Considering the importance of feature extraction in this field, in this paper, we first present and discuss several widely used edge operator, including Sobel, Prewitt, Roberts, Canny and LoG. Furthermore, an experiment is conducted to compare performance and accuracy of five operators by applying them to a leaf image taken from agricultural crop for edge detection. The results of experiment show that, in practice, LoG edge operator is relatively a better choice and performs well for edge detection of agricultural crop leaf image.

This paper studies how to deal with the security issues of agricultural images in remote transmission. An encryption algorithm based on the Logistic and the Henon maps is proposed, which uses chaotic iteration to generate the encryption keys, and then carries out the XOR and cyclic shift operations on the plain text to change the values of image pixels. The algorithm ensures the security of agricultural images during remote transmission.

A new method (RPSAM) to solve global optimization problem is proposed and applied in the process of image registration in agriculture engineering. The random search strategy of simulated annealing is put into the Powell algorithm, and improved by the advancement of the method of selecting starting point. Moreover improving PSAM enable itself possess the characteristic of local quadratic convergence. As the result of experiment turns out, the developed algorithm can prevent optimizing process from trapping into the domain near local minima. Compared with PSAM, RPSAM is improved impressively on the precision of result and efficiency of course in order to improve the speed and quality of image in the course of image registration in agriculture engineering.

In recent years, the mutual information is regarded as a more efficient similarity metrics in the image registration. According to the features of mutual information image registration, the Bee Evolution Genetic Algorithm (BEGA) is chosen for optimizing parameters, which imitates swarm mating. Besides, we try our best adaptively set the initial parameters to improve the BEGA. The programming result shows the wonderful precision of the algorithm.

Ridgelet transform as a time-frequency and multiresolution analysis tool is more powerful than wavelet analysis in the signal and image processing domain, especially in image restoration. Due to the difficulty to appraise the sorts of noise produced by optical imaging equipments inevitably, this paper use independent component analysis to separate the independent signals from overlapping signals. Then ridgelet transform were applied to decompose it, and use a new thresholding de-noising approach to remove noise. At last, we reconstructed the image to obtain a restoration image. By contrast, the efficiency of our method is better than other traditional filtering approaches.

The existing Chinese Minorities OCR system is mainly oriented in the "literacy" level, the script recognition has not attracted the attention it deserves, and the area of recognizing the kinds of Chinese minority scripts is still in a blank. This paper presents a method of recognizing the kinds of Chinese minority scripts based on wavelet analysis and Multinomial Naive Bayes. The method of recognizing the kinds of Chinese minority scripts based on wavelet analysis and Multinomial Naive Bayes is presented which adopts wavelet decomposition that obtains feature descriptor of wavelet energy and wavelet energy distribution proportion. Combined with the texture feature of Chinese minority scripts, radially classification in Multinomial Naive Bayes. Among Chinese, English and Chinese minority scripts such as Tibetan, Tai Lue, Naxi Pictographs, Uighur, Tai Le, Yi, the experimental results show the recognition rate is up to 90%.

Most physiological and environmental changes are capable of inducing variations in animal behavior. The behavioral parameters have the possibility to be measured continuously in-situ by a non-invasive and non-contact approach, and have the potential to be used in the actual productions to predict stress conditions. Most vertebrates tend to live in groups, herds, flocks, shoals, bands, packs of conspecific individuals. Under culture conditions, the livestock or fish are in groups and interact on each other, so the aggregate behavior of the group should be studied rather than that of individuals. This paper presents a method to calculate the movement speed of a group of animal in a enclosure or a tank denoted by body length speed that correspond to group activity using computer vision technique. Frame sequences captured at special time interval were subtracted in pairs after image segmentation and identification. By labeling components caused by object movement in difference frame, the projected area caused by the movement of every object in the capture interval was calculated; this projected area was divided by the projected area of every object in the later frame to get body length moving distance of each object, and further could obtain the relative body length speed. The average speed of all object can well respond to the activity of the group. The group activity of a tilapia (Oreochromis niloticus) school to high (2.65 mg/L) levels of unionized ammonia (UIA) concentration were quantified based on these methods. High UIA level condition elicited a marked increase in school activity at the first hour (P<0.05) exhibiting an avoidance reaction (trying to flee from high UIA condition), and then decreased gradually.

The traditional method of detecting Chironomid larvaes and plankton in water mostly is observation by Naked Eye, which is inefficient and inaccurate. This paper puts forward the Chironomid larvae image recognition method which is based on the support vector machines and multi-layered wavelet decomposition. Gradation histogram balance strengthening treatment is carried out for the image, so as to improve the contrast ratio and make for the threshold division. For each image, a 36 dimension feature vector is computed via two-level discrete Wavelet transform (DWT). The last step of the proposed approach consists is using support vector machine(SVM) as classifer and Wavelet energy as features to recognize the images. Extensive classification experiments on our image data validate that it is promising to employ the proposed texture features to recognize Chironomid larvaes in image.