Search

Objectives: The quality of network structure is the key to obtain excellent classification performance. Most of existing deep models are designed by the trial-and-error method. When the same model processes different hyperspectral images (HSIs), different network hyperparameters need to be set artificially to adapt to the data characteristics of different images, greatly increasing the difficulty of network design. Aiming at the problem of deep learning model design in hyperspectral image classification, this paper proposes an automatic network structure search method for HSI classification. Methods: In the proposed method, the differentiable structure search technology is first used to search the network structure on the source HSI data sets, then the deep network model is constructed by stacking cells, and finally the classification performance is evaluated on the target HSIs. It should be pointed out that, the proposed method only conducts one search on the source HSI data sets, and the obtained deep model can be applied to other target HSI classification tasks, which can effectively improve the utilization rate of the model. To improve the generalization ability and classification accuracy of the model, multi-source and multi-resolution HSIs are employed to construct the precollected data set, and the partial channel connection operation is introduced to further improve the search efficiency. Results: Three HSIs inculding Houston, Chikusei and Pavia Center are selected as source data. Four HSIs including University of Pavia, Indian Pines, Salinas and Houston 2018 are selected as the target HSIs. The source data is used to determine the network model and the target HSIs are used to evaluate the classification performance. Experimental results show that the proposed method can automatically search out a universal deep network model suitable for HSI classification task, and this model can achieve better classification performance than the conventional deep learning models. The overall classification accuracies of University of Pavia, Indian Pines, Salinas and Houston 2018 are 98.15%, 98.74%, 97.30% and 74.47%, respectively. Conclusions: Compared with other deep models, the model in this paper is automatically determined by differentiable architecture search algorithm according to the data characteristics of HSIs, effectively improving the classification accuracy of target HSIs while avoiding complex network design. Extensive experiments verify the applicability and effectiveness of the proposed method in HSI classification. [ABSTRACT FROM AUTHOR]

Since the objects in remote sensing images have the characteristics of arbitrary direction, dense distribution and large scale differences, object detection in remote sensing images has become a challenging problem. Aiming at this problem, this paper systematically reviewed the work related to rotating object detection in deep learning remote sensing images in recent three years. Firstly, this paper introduced the representation method and characteristics of rotated bounding box. Then, it analyzed the existing methods of rotating object detection in remote sensing images from four aspects: the feature extraction network, the generation of rotating anchor and candidate bounding box, the label allocation and sampling strategy, and the loss function. Next, it introduced the commonly used remote sensing image datasets for rotated object detection, and compared the performance of different algorithms. Finally, this paper prospected rotating object detection in remote sensing images. [ABSTRACT FROM AUTHOR]

Recognizing the phase transition of matter is an important problem in physics research. Convolutional neural network algorithm of confusion label scheme is used to identify quantum phase transition point of two- component Bose-Einstein condensates (BEC) in this paper. By calculating the output accuracy of neural net- work, the W-shape performance curve is obtained. The maximum value in the middle of W-shape performance curve corresponds to the critical point of quantum phase transition. The research results show that the critical point obtained by deep learning is consistent with the analytic calculation results. The deep learning method of confusion label scheme can be applied to the phase transition system of existing two phases. [ABSTRACT FROM AUTHOR]

Although high-quality and high-resolution depth maps can significantly improve the performance of computer vision tasks in various natural scenes, the limitations of depth camera hardware make the depth maps captured by consumer-level depth cameras have problems such as low resolution, poor quality, invalid holes, etc. Depth super-resolution reconstruction (DSR) is a kind of technology that can effectively improve the resolution and quality of depth maps, and DSR has become a research hot-spot in the field of computer vision. Firstly, this paper introduced the definition of DSR and the research progress of DSR algorithm at home and abroad in recent years, and then mainly stated and analyzed deep learning DSR reconstruction algorithms. Next, it introduced the depth image quality evaluation criteria. Finally, this paper prospected the application fields of DSR and the challenges as well as the opportunities in the future. [ABSTRACT FROM AUTHOR]

As an important research direction in the field of machine learning, deep learning technology has been infiltrated into various fields of scientific research due to its powerful feature of learning ability and excellent portability. In particular, the achievements in the frontier fields of deep learning have played a positive role in the recognition, classification, segmentation and quantification of medical images. Optical coherence tomography (OCT) technology has the characteristics of non-invasive, high resolution, non-invasive acquisition of micron-level high resolution 3D sectional image of biological tissue, and has been widely used in ophthalmology, vascular, dermatology and other biomedical fields. This paper first introduces the basic principles of OCT and deep learning and the advantages of their combination in the field of skin, and then elaborates the application of deep learning in the field of skin OCT medical images, including the analysis and classification of skin cancer, beauty, fingertip image segmentation, and the qualitative and quantitative evaluation of damaged skin. [ABSTRACT FROM AUTHOR]

Expert recommendation is one of the research hotspots in the online Q&A community. However, the majority of the currently used algorithms concentrate on matching user's static interest and question information, ignoring the efficient capture of user' s dynamic interest representation information, which results in a lack of recommendation accuracy. In order to address the aforementioned issues, this paper proposed an expert recommendation algorithm for the Q&A community that incorporated multi-head self-attention. Firstly, this algorithm built out question encoders of convolutional neural networks and attention mechanisms to process target question and user' s historically answered questions, in order to extract the corresponding question representations. Secondly, it treated the sequence of user's historically answered questions as a time series and used the multi-head self-attention mechanism to learn the dynamic interest representations hidden in the sequence, and combined with the user' s static interest representations to obtain the user' s comprehensive interest representations. Finally, it calculated the similarity between the target question representation and the comprehensive user representation to generate recommendation results. Based on the real data from the Zhihu Q&A community, the experimental results show that the performance of this algorithm is significantly better than the current popular expert recommendation algorithm. [ABSTRACT FROM AUTHOR]

China is the major producer and consumer of pork all around the world. In China, pork production has accounted for more than 60% of meat production for a long time. Nowadays, the pig industry has gradually shifted from decentralized and extensive breeding to intensive large-scale breeding, promoting production efficiency. At present, as people's demand for the quantity and quality of pork is increasing, China's pig industry faces the problem that the output and quality of pork can not meet people's daily needs. With the rise of artificial intelligence technology in recent years, deep learning technology has been developed rapidly and widely used in image and audio recognition, natural language processing, robotics, bioinformatics, chemistry, finance, and other fields. It is also an essential tool in developing precision livestock farming. The body condition, behavior and health status of pigs could directly affect the income level of pig farms, so through using deep learning technology, we can quickly and accurately acquire the relevant information about pigs, and carry out precise management to improve the feeding efficiency and welfare levels of pigs. This paper expounds on the research progress and application status of deep learning technology used in target pig detection, pig image segmentation, pig body condition and abnormal monitoring, as well as pig behavior recognition. Then we put forward the improvement strategy of deep learning technology used in the process of pig feeding, which make it easier for researchers to understand. At the same time, we summarize and analyze the data sources and datasets, application scope, and model optimization of previous works using deep learning technology in pig breeding. In the field of data sources and datasets, mobile computer vision systems are currently more suitable than systems with many fixed cameras when they are applied in pig houses, therefore, further research could focus on how to use deep learning technology to mobile computer vision systems. Deep learning technology requires a large amount of data to learn data features, which is a massive disadvantage for pig applications. To get pre-training weights suitable for agricultural data sets, a large number of public datasets and a unified dataset standard suited to the pig field should be established. In terms of application scope, due to the short application time of deep learning in the process of raising pigs, many critical occasions are not involved or seldom involved in existing research, so the application scope of deep learning should further expand. In model optimization, optimizing the model to meet the needs of practical application scenarios is the direction of future research. Optimizing the model to achieve a better balance between size and model performance requires further study. Optimizing the model to locate the start and end time of the behavior in unclipped video and recognize the behavior of target pigs is the challenge in identifying the behavior of pigs in future. China's research on deep learning is at the top level, so the application prospect of deep learning in pig farming is highly expected. Combined with the actual production scene of pigs, deep learning technology will significantly contribute to the development of precision livestock farming. [ABSTRACT FROM AUTHOR]

To solve the problems that the objects with different semantic classes are identified as co-salient objects in current co-saliency detection methods, this paper proposed a CNN and semantic correction-based co-saliency detection method (CSCCD). The proposed method first adopted the guided super pixel filter to process the super pixels obtained by SLIC and the saliency results generated by DSS, which showed clear object boundaries. Then it utilized Mask R-CNN to extract semantic features. It proposed the definitions of image semantic feature and semantic consistency. It also defined the image group semantic correction to solve the problem that detected the objects with different pose belonging to a semantic class as different semantic classes. With the concept, this paper defined image group semantic correlation class, solving semantic correlation problem of multiple images. It generated the final co-saliency detection results by fusing the saliency detection regions with the image group semantic consistent regions. The experimental results on public benchmark datasets show that this algorithm can effectively highlight the whole and outline of the object, and its comprehensive performance in objective quantification is obviously improved. [ABSTRACT FROM AUTHOR]

Accurate detection and recognition of plant diseases is the key technology to early diagnosis and intelligent monitoring of plant diseases, and is the core of accurate control and information management of plant diseases and insect pests. Deep learning can overcome the disadvantages of traditional diagnosis methods and greatly improve the accuracy of diseases detection and recognition, and has attracted a lot of attention of researchers. This paper collected the main public plant diseases image data sets all over the world, and briefly introduced the basic information of each data set and their websites, which is convenient to download and use. And then, the application of deep learning in plant disease detection and recognition in recent years was systematically reviewed. Plant disease target detection is the premise of accurate classification and recognition of plant disease and evaluation of disease hazard level. It is also the key to accurately locate plant disease area and guide spray device of plant protection equipment to spray drug on target. Plant disease recognition refers to the processing, analysis and understanding of disease images to identify different kinds of disease objects, which is the main basis for the timely and effective prevention and control of plant diseases. The research progress in early disease detection and recognition algorithm was expounded based on depth of learning research, as well as the advantages and existing problems of various algorithms were described. It can be seen from this review that the detection and recognition algorithm based on deep learning is superior to the traditional detection and recognition algorithm in all aspects. Based on the investigation of research results, it was pointed out that the illumination, sheltering, complex background, different disorders with similar symptoms, different changes of disease symptoms in different periods, and overlapping coexistence of multiple diseases were the main challenges for the detection and recognition of plant diseases. At the same time, the establishment of a large-scale and more complex data set that meets the specific research needs is also a difficulty that need to face together. And at further, we point out that the combination of the better performance of the neural network, large-scale data set and agriculture theoretical basis is a major trend of the development of the future. It is also pointed out that multimodal data can be used to identify early plant diseases, which is also one of the future development direction. [ABSTRACT FROM AUTHOR]

Because convolutional neural networks( CNN) mostly focus on global feature learning and ignore local feature information containing more details, it is difficult to improve the accuracy of indoor scene recognition. To solve this problem, this paper proposed an indoor scene recognition method based on improved global-local attention network ( GLANet). Firstly, it used GIAN et to capture the global and local features of the scene image to increase the detail information of the image features. Then, it introduced a non-local attention module into the local network to further preserve the detail features of the image through the convolution of attention diagram and feature graph. Finally, it used various features of different stages of the network for classification. Through training and verification on the MIT lndoor67 and the SUN397 dataset, the recognition accuracy of the proposed method was increased by l. 98% and 3. 07% respectively compared with the LGN method. Experimental results show that the algorithm can effectively capture global semantic information and fine spatial details, and significantly improves the recognition accuracy. [ABSTRACT FROM AUTHOR]

Aiming at the knowledge overload in the process of online learning and the problems of data sparsity and cold start in the traditional recommendation algorithm, this paper proposed an improved learning resource recommendation algorithm based on MLP. The algorithm used the advantage of MLP in nonlinear data processing, converted the prediction mode of vector multiplication of learner characteristics and learning resource characteristics into the input mode of MLP, and improved the DN-CBR neural network recommendation model. In order to verify the effectiveness of the model, this paper built a dataset with the online learning platform data of iCourse as the sample. Compared with DN-CBR model, the improved model improves the normalized cumulative loss gain and hit rate by 1. 2 % and 3 % respectively on the dataset, which effectively improves the recommendation performance of the model. [ABSTRACT FROM AUTHOR]

Alzheimers disease (AD) is an irreversible neuro degenerative brain disease and the most common dementia in the elderly. Manual classification of Alzheimer's magnetic resonance image (MRI) has problems delay classification and time-consuming classification. As the aging population becomes more and more serious accurately and quickly classify patients with AD has important research significance. This paper combines convolutional neural network (CNN) technology with MRI technology and designs a 3D-ResNet model for AD classification which achieves 98. 39% accuracy 96.74% sensitivity and 99. 99% specificity on the validation set and achieves 97. 43% accuracy 94. 92% sensitivity and 99. 99% specificity on the test set. The classification time of each patient is 0. 23 s. In addition，for the problem that the pathogenesis of AD is not yet clear，this paper uses Class Activation Mapping (CAM) technology to visualize the AD-related brain regions. [ABSTRACT FROM AUTHOR]

Firstly, this paper reviewed the development of computer vision, presented the elementary knowledge of neurons, multi-layer perceptron, back-propagation algorithm, and the development of convolutional neural network (CNN) and its basic operation of convolution and pooled etc. Secondly, it discussed the structures of typical convolutional neural networks such as AlexNet, VGGNet, GoogLeNet and ResNet etc., and mainly introduced CapsNet. And then it summaried the research progress of the application of CNN in image classification, image semantic segmentation, image target detection and image generation. At last, this paper presented the research challenge of CNN and the future next work for CapsNet. [ABSTRACT FROM AUTHOR]

Pigs like to fight with each other to form a hierarchy relationship in groups. Aggressive behaviors, mostly fighting, are frequently found in intensive pig raising facilities. Strong aggressive behaviors can cause other pigs lack of food and water, growing slowly, wounds, sick and even dead in serious situation. This considerably reduces health and welfare of pigs and further decreases economic benefits of pig industries. Monitoring and recognizing aggressive behaviors among pig group is the first step to manage the aggressive behaviors in group pigs effectively. Traditional human recording method is time-consuming and labor-intensive. This method can’t be used 24 hours a day, 7 days a week. Machine vision technique brings an automatic monitoring method to solve this problem. In this paper, we introduced a new method for aggressive behaviors monitoring based on deep learning. The experiments were held under controlled environments, which were achieved in an environment-controlled chamber designed previously. The details of the chamber were depicted in a published paper written by our research group. Nursery pigs were fed under three different concentration levels of NH3 gas, which were <3.80 mg/m³, 15.18 mg/m³, 37.95 mg/m³, with a suitable temperature of around 27 ℃ and the comfortable humidity between 50%-70%. Each nursery group had six pigs and were weight around 9.6 kg. During each 28 days’ experiment of three concentration levels of NH3, videos were taken from the top of the chamber. An end-to-end network, named 3D CONVNet, was proposed for aggressive behavior recognition of group pigs in this paper, which based on a C3D network and built with 3D convolution kernels. The network structure of the 3D CONVNet was improved in both width and depth dimensions. The number of main convolutional layers was increased to 19, extra batch normalization and dropout layers were added to deepen the network. Furthermore, the multi-scale feature fusion method was introduced to widen the network. This improvement had bettered the performance of the algorithm considerably. To train the 3D CONVNet, 380 aggressive (14 074 frames) and 360 none-aggressive videos (13 040 frames) were chosen from experimental videos recording in experiments of two concertation levels. These videos were randomly divided into training set and validation set, and the ratio of each set is 3:1. Another 556 aggressive videos and 510 none-aggressive videos from the three experimental batches were chosen to build the testing set. There was no overlap among training set, validation set, and testing set. Results showed a total of 981 videos, including aggressive and non-aggressive behaviors, was correctly recognized from the whole 1066 testing videos. The precision of the 3D CONVNet was proved to be 92.03% on testing set. Among them, the precision, recall rate and F1-Score for aggressive behaviors were 94.86%, 89.57%, and 92.14%, respectively. The precision for different NH3 concentration experimental levels were 94.29%, 89.44%, and 85.91%, respectively, which showed the generalization performance of the 3D CONVNet. With the similar heat environments, the 3D CONVNet also showed the good performances under different illumination condition. The comparison with C3D，C3D_1 (19 layers) and C3D_2 (BN) networks resulted in 95.7% on validation set, 43.27 percent higher than the C3D network. The recognition on single image using the 3D CONVNet was only 0.5 s, which was much faster than the other three networks. Therefore, the 3D CONVNet was effective and robust in aggressive behavior recognition among group pigs. The algorithm provides a new method and technique for aggressive behavior auto-monitoring of group pigs and helps improve establishment of auto-monitoring system in pig farms and manage level of pig industry. [ABSTRACT FROM AUTHOR]

For the traditional objective function based on the minimum classification error criterion ( MCE ), there exists the problem of gradient inversion in the network when the sample is misclassified. This paper introduced the minimum classification error criterion and defined the FMCE objective function with correction term. This paper used the higher precision cross entropy as a basis function and FMCE as a correction function, and proposed an improved cross entropy objective function CE-FMCE, which made the network increase the probability of label class output in the reverse propagation process. CE-FMCE not only overcame the gradient inversion problem of the traditional MCE objective function, but also compensated for the lack of differentiation of the cross entropy function for non-label set gradients. This paper compared CE-FMCE with MSE, cross entropy, MCE, and M3 CE on self-built typhoon image dataset and general dataset MNIST. The experimental results show that CE-FM CE is superior to other objective functions. [ABSTRACT FROM AUTHOR]

At present, pattern recognition technology has been widely used in the fields of objects, faces, fingerprints, military target recognition, etc. However， pattern recognition method still has obvious short⁃ comings when applied to the above fields. It is currently restricted to the use of image information for identi⁃ fication. When the image features of the research object are highly similar, the accuracy of pattern recogni⁃ tion is low and cannot meet the actual application requirements. For example, in the case of mixed true and false targets, it is difficult to obtain satisfactory recognition results using only image information. Aiming at the above problems, a pattern recognition method integrating image information and spectral information is proposed in this paper. Firstly, the image recognition model based on the convolutional neural network model is built to identify object categories based on the semantic features of objects and obtain preliminary recognition results. Then, on the basis of the preliminary recognition results, the measured spectral data of the object (spectrum range 400-1 000 nm, spectral resolution 2 nm) is used to perform true and fake identifi⁃ cation of the object based on the back propagation (BP) neural network model. The principle of true and false recognition is that the true and false targets are different in material, causing significant difference in their hyperspectral information. Finally, recognition results are obtained. In order to verify the accuracy of the proposed method, true and fake apples and grapes are used as experimental subjects and the result is that: The recognition accuracy obtained by using only image information is 38.50%, and the recognition accuracy obtained by using only spectral information is 63.00%, however, the recognition accuracy obtained by the method proposed in this paper is 95.00%. Compared with the existing pattern recognition method without spectral information participation, the pattern recognition method using image information and spectral information proposed in this paper improves the pattern recognition accuracy under the mixed condi⁃ tion of true and false targets, and can be widely applied to object recognition, face recognition, fingerprint recognition, military target recognition and other fields. [ABSTRACT FROM AUTHOR]

Aiming at the problem that the specific classification of geometric surfaces on the set of image recognition algorithms could lead to the loss of discriminative information, this paper proposed an improved iterative depth learning algorithm ( 1113-LA ) based on convolution neural network(CNN). The algorithm used the hybrid convolution network to perform the translation invariant feature learning at the bottom, and iteratively used CNN to learn different nonlinear characteristics of the input image set in a hierarchical way. The graphic library and query examples of the algorithm included different perspectives, backgrounds, facial expressions, resolution and illumination, and the image sets of objects or objects. This paper compared the proposed algorithm with other algorithms on a data set. The experimental results show that the proposed algorithm has the best performance on the measured data set. [ABSTRACT FROM AUTHOR]

Mechanical weeding can reduce the use of pesticides and is of great significance to ensure high yield of crops. Real-time and accurate identification of crops is a key technical problem needs to be solved in mechanical weeding equipment. Because of the subjectivity of feature extraction process in weed recognition, the accuracy of traditional methods in actual field environment is low. In recent years, the method of weed identification based on convolution neural network has been widely studied. Although the accuracy is obviously improved, there are still problems such as large parameters and poor real-time performance. In order to solve the above problems, a four-channel input image is constructed by collecting near infrared and visible images of sugar beet in the field, and a lightweight convolution neural network based on codec structure is proposed. In this paper, Sugarbeet and weed images collected from a farm in Bonn, Germany, in 2016 were used as data sets, which covered images of different growth stages of sugar beet, and 226 pictures of which were randomly selected as training sets, and the remaining 57 pictures were used as test sets. The experimental data set was composed of three channels of visible light image and one channel of near infrared image, which are merged into a four-channel image by pixel level superposition, and the depth-wise separable convolution was used in the deep model. Firstly, the input feature image was convoluted in 2 dimensions convolution kernel and the number of channels was expanded. Then, the 1×1 convolution kernel was used to make the 3 dimensions convolution which combined channel feature and compressed the channels to enhance the nonlinear mapping ability of the model. In order to avoid the problem of the gradient disappearing, the residual block was used to connect the input and output of the depth-wise separate convolution. Finally, the coding and decoder structure was designed and the shallow features were combined with deep features to refine the segmentation effect. Due to the imbalance of pixel proportions of soil, crops and weeds, the weighted loss function was used to optimize the model. The segmentation accuracy, parameters and operating efficiency of models at different input resolutions and different width factor were introduced to evalute the model. When the width factor was 1, the segmentation accuracy of the model increased with the increase of the input image resolution, the model accuracy of four channel input was higher than that of the model based on original visible image input, which showed that the near-infrared image features can compensate the defects of ordinary RGB images to some extent, and make the model more suitable for the dark environment. Under the same input image resolution, the model with a width factor of 2 or 4 performs better than the model with a width factor of 1. With the increases of width factor, the parameters of the model increase greatly. The amount of calculation is related to the size of the input image, so the frame rate gradually decreases with the increase of the size of input image. The experimental results show that the optimal model in this paper is a four channel input model with a width coefficient of 2, and the average intersection union ration is 87.58%, the average pixel accuracy is 99.19%, the parameters are 525 763 and the frame rate is 42.064 frames/s. The model has high segmentation and recognition accuracy and good real-time performance, and can provide theoretical basis for the development of intelligent mechanization weeding equipment. [ABSTRACT FROM AUTHOR]

Due to the small amount of microscopic features image data of traditional Chinese medicinal materials powder, unbalanced distribution of sample classes and small difference between classes, it is difficult to achieve a satisfying classification effect through traditional image classification methods. To solve the above problems,this paper proposed an improved method of deep convolution neural network based on dynamic ReLU and attention mechanism model. Firstly, it used Xception as the basic network, which had an obvious effect on small sample data classification. Secondly, it replaced the static ReLU activation function in the network with the improved dynamic ReLU function, so that each sample had its own unique ReLU parameters. Finally, it embedded the improved SE module in the network to enable the network to learn the importance of each feature channel automatically. The proposed method can make the network pay more attention to the detailed information in the image, and can solve the problem of unbalanced distribution of sample classes and small differences between classes. The experimental results show that the image classification accuracy of 56 kinds of traditional Chinese medicinal materials powder vessel is increased by about 1.5% to 93.8%, which demonstrates that the proposed method is advantageous over other image classification methods. [ABSTRACT FROM AUTHOR]

Face makeup transfer refers to the task of transferring reference makeup to people’s face while keeping the features of people’s face and showing the style of the reference makeup as much as possible. The current methods of transferring facial makeup do not fully consider the facial differences,it causes the problems such as insufficient extracted facial information. In order to avoid such problems and realize automatic transferring of face makeup,this paper proposed a face makeup transfer algorithm based on deep convolutional neural network. First,the algorithm automatically located the people’s face and features of reference makeup,and extracted the feature information of key parts. After the autonomous training of deep convolutional neural network,it automatically transferred the reference makeup to people’s face through the makeup transfer network and loss function. The simulation result shows that the algorithm takes less time,has more advantages in computing performance,and makeup transfer effect is more natural without changing the details of the original facial features. [ABSTRACT FROM AUTHOR]

Object detection for seafood is a key to intelligence marine ranching, but its real-time and accuracy need to be improved. In this paper we present an YOLOv3-based real-time object detection method for seafood. Firstly, for better real-time performance, we replace the 3×3 standard convolution in YOLOv3 with the depthwise separable convolution technology. This technology resolves standard convolution into depthwise convolution and pointwise convolution. It greatly reduces the amount of parameters and calculations of the model. Thus, we obtain a lightweight network model, named Depthwise Separable Convolution-YOLO (DSC-YOLO); Then, in the data preprocessing stage, we introduce UGAN-based image enhancement methods to relief the color distortion of underwater seafood images. This method uses an unsupervised method to automatically learn the distribution of training data and generate new data. The network consists of generator network and discriminators network. Its main idea is confronting and gaming between networks, on the basis of original data information to generate a new data information. It realizes the mapping process from underwater blurred image to underwater clear image. To be specific, we use CycleGAN to convert images in air to underwater-like images. Thus, lots of image pairs can be obtained and used to train UGAN model. Once UGAN finish training, learned generator of UGAN can transform real underwater images to clear images. Finally, we introduce Mosaic-based data augment methods to enrich image diversity. To be specific, four images are randomly selected from underwater dataset images, and operated with traditional data augmentation (rotation, translation, scaling, roll). Then, the four images are randomly cropped and spliced, obtaining a Mosaic image. The new image has the same dimension as the original images. Combining the three strategies, we obtain a real-time object detection method for seafood. In order to verify the effectiveness of our proposed method, we conduct several experiments. In these experiments, DSC-YOLO, UGAN underwater image enhancement and Mosaic data enhancement were combined to obtain four models: DSC-YOLO, DSC-YOLO+UGAN, DSC-YOLO+Mosaic, DSC-YOLO+UGAN+Mosaic. The four models compare with the YOLOv3 model over the same dataset. The results of experiments show that compared with YOLOv3 model, the model size of the proposed method reduces by 70%, the inference time reduces by 16%, the Recall (R) improves by 2.7%, the mean Average Precision 0.5 (mAP0.5) promotes by 2.4%, and the F1-Score increases by 0.4%. The results indicate that the proposed method has the potential to meet real-time requirements and deploy on mobile devices. [ABSTRACT FROM AUTHOR]

Mechanical weeding can reduce the use of pesticides and is of great significance to ensure high yield of crops. Real-time and accurate identification of crops is a key technical problem needs to be solved in mechanical weeding equipment. Because of the subjectivity of feature extraction process in weed recognition, the accuracy of traditional methods in actual field environment is low. In recent years, the method of weed identification based on convolution neural network has been widely studied. Although the accuracy is obviously improved, there are still problems such as large parameters and poor real-time performance. In order to solve the above problems, a four-channel input image is constructed by collecting near infrared and visible images of sugar beet in the field, and a lightweight convolution neural network based on codec structure is proposed. In this paper, Sugarbeet and weed images collected from a farm in Bonn, Germany, in 2016 were used as data sets, which covered images of different growth stages of sugar beet, and 226 pictures of which were randomly selected as training sets, and the remaining 57 pictures were used as test sets. The experimental data set was composed of three channels of visible light image and one channel of near infrared image, which are merged into a four-channel image by pixel level superposition, and the depth-wise separable convolution was used in the deep model. Firstly, the input feature image was convoluted in 2 dimensions convolution kernel and the number of channels was expanded. Then, the 1×1 convolution kernel was used to make the 3 dimensions convolution which combined channel feature and compressed the channels to enhance the nonlinear mapping ability of the model. In order to avoid the problem of the gradient disappearing, the residual block was used to connect the input and output of the depth-wise separate convolution. Finally, the coding and decoder structure was designed and the shallow features were combined with deep features to refine the segmentation effect. Due to the imbalance of pixel proportions of soil, crops and weeds, the weighted loss function was used to optimize the model. The segmentation accuracy, parameters and operating efficiency of models at different input resolutions and different width factor were introduced to evalute the model. When the width factor was 1, the segmentation accuracy of the model increased with the increase of the input image resolution, the model accuracy of four channel input was higher than that of the model based on original visible image input, which showed that the near-infrared image features can compensate the defects of ordinary RGB images to some extent, and make the model more suitable for the dark environment. Under the same input image resolution, the model with a width factor of 2 or 4 performs better than the model with a width factor of 1. With the increases of width factor, the parameters of the model increase greatly. The amount of calculation is related to the size of the input image, so the frame rate gradually decreases with the increase of the size of input image. The experimental results show that the optimal model in this paper is a four channel input model with a width coefficient of 2, and the average intersection union ration is 87.58%, the average pixel accuracy is 99.19%, the parameters are 525 763 and the frame rate is 42.064 frames/s. The model has high segmentation and recognition accuracy and good real-time performance, and can provide theoretical basis for the development of intelligent mechanization weeding equipment. [ABSTRACT FROM AUTHOR]

As an experimental technique for fluid mechanics, particle image velocimetry (PIV) can extract global and quantitative velocity field from images. With the development of artificial intelligence, designing PIV method based on deep learning is quite promising and worth studying. First, the authors in this paper introduce the optical flow neural network which is proposed in the computer vision community. Second, a dataset including particle images and the ground-truth fluid motions is generated to train the parameters of the networks. This leads to a deep neural network for particle image velocimetry which can provide dense motion estimation (one vector for one pixel) efficiently. The featuring of particle image extracted by the neural network is also investigated in this paper. It is found that feature matching improves the accuracy of estimation. The proposed network model is firstly evaluated by a synthetic image sequence of turbulent flow. A jet flow experiment is also carried out in this paper to validate the practicability. The experimental results indicate that compared with the traditional cross-correlation and optical flow methods, the proposed deep neural network has advantages in accuracy, spatial resolution as well as efficiency. [ABSTRACT FROM AUTHOR]

Aiming at the problem of time-consuming,complex feature extraction and low recognition rate,this paper introduced the convolution neural network method based on deep learning. This method was widely used in the field of image recognition because of its good robustness, strong generalization ability and high recognition rate. In classification experiments of the 4 main vehicle types (bus, microbus, car, truck) on road, the improved convolution neural network LeNet-5 achieved vehicle type training and test accuracy both over 98% , which outperformed traditional SIFT -I- SVM algorithm. In addition, this paper also studied the effect of the Dropout layer in improved network on the vehicle type recognition. Compared with the traditional algorithm, the improved convolution neural network LeNet-5 has a significant improvement in reducing the detection time and improving the recognition rate,and has obvious advantages in vehicle type recognition. [ABSTRACT FROM AUTHOR]

In the recommendation system. the inner product interaction of a single learning matrix de composition or the use of deep neural networks to capture user interaction with the project is not sufficient to effectively learn the potential characteristics of users and projects. In view of this problem. this paper proposes a collaborative filtering algorithm based on deep learning feature representation (DLFeaCF) based on display feedback and implicit feedback. The model first learns the inner product and outer product interaction between the user and the project. Then, based on the inner product, it uses the nonlinear interactive learning ability of the multi-layer perceptron (MLP) to obtain the two aspects from implicit mapping and feature mapping: on the basis of the outer product, CNN learning is used to capture the local features of the user and the project finally, combine the features in the fusion layer and obtain the prediction score, The experiments on the real MovieLens dataset show that the DLFeaCF model can achieve better recommendation performance. [ABSTRACT FROM AUTHOR]

Aiming at the problems that the existing human motion recognition methods based on deep learning have large parameters and the networks were too deep and heavy, this paper proposed a lightweight two-steam fusion deep neural network model and applied this model to human action recognition. This model combined a shallow multi-scale network with a deep network, and achieved a significant reduction in the amount of model parameters and avoided the problem that network was too deep. Experiments were performed on datasets UCFlOl and HMDB51, achieving 94.0% and 69.4% recognition accuracy in ImageNet pre-training mode, respectively. Experiments show that compared with the existing human motion recognition models based on deep learning, this model greatly reduces the parameter quantity and still has high motion recognition accuracy. [ABSTRACT FROM AUTHOR]

Effective recognition method of crop and weed is the basis for promoting the development of intelligent mechanization weeding pattern. Summarizing the previous research, we found that the accuracy and stability of the recognition model could be improved by natural and diversified feature presentation, but there are still 2 main problems. On the one hand, feature presentation of the natural property of target was difficult to be obtained by the hand-engineered feature extractor. The spatial consistency of the obtained features was bad, and the real-time performance of recognition system was reduced for the complex feature extraction algorithm. On the other hand, the effect of image preprocessing has important influence on recognition results, especially the overlapping segmentation of crop and weed. For overlapped objects, it is usually difficult to segment them without affecting their respective feature presentations, resulting in low recognition accuracy and stability. In order to solve the main problems in the current research, we explored the way to improve the recognition accuracy, stability and real-time performance, and a recognition method of crop and weed based on multiscale hierarchical feature learning combined with superpixels segmentation was proposed. The main research contents of this paper were as follows: 1) Excellent internal features of image are hierarchical. In this research, the multiscale hierarchical feature is a scene level feature with invariance and consistency in scale space. Multiscale convolutional neural network was built to extract multiscale hierarchical feature. Multiscale convolutional neural network contains multiple copies of a single CNN (convolutional neural network) that are applied to multi-scale Gaussian pyramid of the input image. The CNN model as feature extractor in this paper includes 3 stages. In the first 2 stages, it contains a bank of filters (convolution kernels) to produce dense feature maps, a point to point nonlinear mapping activation function, and a spatial pooling module for sub-sampling of each feature map, While the last stage only contains a bank of filters. Each filter is applied to the input feature maps through 2-dimensional convolution operation, in which local feature presentations are detected at all pixel locations on the input image. For each pixel, the CNN model is used to collectively encode the internal information in a large sense region around the location of given pixel. The CNN is fed with raw pixels and trained with back propagation method. With complete training, this CNN model can automatically extract hierarchical feature representations from the input image, thereby decreasing the need for hand-engineered features extracting. A series of feature maps for multiscale regions centered on each pixel of the image are produced by the multiscale convolutional neural network; these representations contain shape, texture and sense information. Therefore, the multiscale hierarchical feature is learned to allow the recognition of the class of all pixels in the image. The average pixel recognition rate is 93.41%. 2) In this research, multiscale hierarchical feature is used to produce the class distribution for every pixel through a 2-layer MLP (multi-layer perceptron). But recognizing the class of pixel independently from its surrounding regions could produce some interference at the boundary of targets. Accurate boundary segmentation for each target in the image is not provided. In this paper, the superpixel method is used for generating an over segmentation of the image, where each segmentation component (superpixel) is an irregular pixel block composed of similar pixels in texture, color and brightness characteristics. Object in the image is composed of these superpixels, and has the exact original boundary. The superpixels segmentation is computed following the simple linear iterative clustering method described in this paper. 3) We proposed the recognition strategy of multiscale hierarchical feature learning combined with the superpixels segmentation. Firstly, an over segmentation of the original image was produced through the superpixels segmentation. At the same time, each pixel location of the image was classified densely based on the multiscale hierarchical features. These predictions of pixels in each superpixel were aggregated to produce the class prediction of superpixel, through computing the average class distribution within the superpixel. Adjacent superpixels with the same class were merged to obtain the final target class prediction and image segmentation. The accurate image segmentation was achieved while recognizing the target in the image by this recognition method, which effectively avoided the problems caused by targets overlapping, and the recognition results were more stable and accurate. Maize seedlings at 2-5 leaves stage and weed during the same stage were used as research object, and the recognition method of multiscale hierarchical feature learning combined with superpixels segmentation was tested. The results showed that the average target recognition rate with this method was 98.92%, and the standard deviation was 0.55%. The average target recognition rate with the method in previous research was 98.36%, and the standard deviation was 1.05%. So, the accuracy and stability of recognition results by this method in this paper were improved with different degrees. In aspect of real-time performance, the average time to recognize a single image was 1.68 s with this method, and compared with the method proposed in previous research, it was decreased by 1.58 s. The real-time performance of this method can be further improved by GPU (graphics processing unit) hardware acceleration, and the average time to recognize a single image was only 0.72 s. Therefore, the recognition method based on multiscale hierarchical feature learning combined with superpixels segmentation can effectively achieve accurate, stable and efficient recognition of maize and weed. The research results provide reference for the development of precision weeding. [ABSTRACT FROM AUTHOR]

Aiming at the problem that the existing target detection framework based on big data and deep learning is difficult to realize real-time video target detection on low-power mobile and embedded devices, this paper improved the target detection framework SSD based on deep learning, and put forward an improved multi-target detection framework LSTM-SSD which was dedicated to multi-target detection of traffic scenes video. Combining single image detection frame with recursive neural network LSTM network to form an interleaved circular convolution structure, it realized the temporal association of network framelevel information by extracting the feature map between propagation frames by adopting a Bottleneck-LSTM layer, which greatly reduced the network calculation cost. Combining the time-aware information with the improved dynamic Kalman filtering algorithm, the tracking and identification of the targets which were influenced by strong interference such as light change and largearea occlusion in the video could be realized. Experimental results show that the improved LSTM-SSD can achieve good results when dealing with the difficult detection situations such as multi-targets, cluttered background, light changes, fuzziness and large-area occlusion. Compared with other target detection frameworks based on deep learning, the average accuracy rate of all kinds of target identification is increased by 5% - 16%, the average accuracy rate is increased by 4% - 10%, the multi-target detection rate is increased by 4% - 19%, and the detection frame rate reaches 43 fps, basically meeting the requirements of real-time . The balance between the accuracy of the algorithm and the running speed is achieved, and a good detection and identification effect is achieved. [ABSTRACT FROM AUTHOR]

The problem that the recognition rate of handwritten Chinese characters is affected by random noise, this paper proposed a new algorithm based on deep learning and noise suppression. This algorithm was mainly aimed at handwritten Chinese character characters and pictures with random noise. It was a model that used Caffe platform to establish noise suppression and convolutional neural networks in the Python environment. It removed noise and correctly recognized handwritten Chinese characters. In addition, the new algorithm did not change the character while removing noise, and retained the original Chinese character information. As a result, the noise intensity of two different types of noise (Gaussian noise and salt-and-pepper noise) was gradually increased, it performed multiple experiments and compared them with other methods, the average recognition rate was 97. 05%. The experimental results show that the model and algorithm have the advantages of high efficiency and strong recognition ability. [ABSTRACT FROM AUTHOR]

Cotton (Gossypium hirsutum) is one of the most important cash crops in China, The timely and effective removal of weeds in cotton seedling stage is an important measure to ensure high and stable yield of cotton. Nowadays, weed recognition based on machine vision is widely used. The fast and effective recognition of crop and weed in the field under natural illumination is one of the key technologies for the development of intelligent mechanization weeding pattern. In the one hand, cotton and weeds have similar color feature in the field. Feature presentation of the natural property of target is difficult to be obtained by the hand-engineered feature extractor. The spatial consistency of the obtained features is not good, and the real-time performance of recognition system is reduced for the complex feature extraction algorithm. On the other hand, the effect of image preprocessing has important influence on recognition results. In order to solve the main problems in the current research, we explored the way to improve the recognition accuracy, stability and real-time performance, and a recognition method of crop and weed based on Faster R-CNN．In this paper, cotton seedling at 2-5 leaves stages and weeding during the same stage were used as research objects under natural illumination. Weed identification from digital images taken under natural illumination at field level is still challenging in agricultural image processing applications, though a lot of research has been conducted related to this topic. To address this problem, images including cottons and weeds were taken vertically from top to bottom. A method based on Faster R-CNN convolutional neural network was proposed to identify weeds from cotton plants more accurately and quickly. The residual network was used to extract image features, with ReLU as the activation function and Max-pooling as the down-sampling method. In the region of proposal network, feature pyramid network was introduced to generate target candidate frame, and Softmax regression classifier was utilized to optimize the CNN network. The proposed methodology was implemented on 200 digital images taken under natural illumination. The experimental results demonstrated that, the average accuracy of weed identification reached 95.5%, and the average time for individual weed plant identification was 1.51 s, which was reduced to 0.09 s by using GPU. To test the efficiency of the proposed methodology, YOLOv3 method was also carried out on the same training and test datasets. The weed identification results were assessed by mean average precision and average precision. The experimental results showed that better performance was achieved by using our proposed methodology, and better identification accuracy was reached as well. This indicated that the proposed method had a good effect on weed detection under natural illumination, and it will greatly promote the development of precise weed control. [ABSTRACT FROM AUTHOR]

Aiming at the problem that the existing human motion recognition method needed to input a fixed length video segment and underutilized the spatio-temporal information, this paper proposed a deep neural network model based on the combination of space-time pyramid and attention mechanism. This improved architecture combined 3D-CNN including spatio-temporal pyramids with LSTM model with spatio-temporal attention mechanism, and realized multi-scale processing of video segments and full utilization of complex spatio-temporal information of actions. For the architecture, the inputs of spatial and temporal domain were RGB image and the optical flow, the input of the fusion domain was the fusion feature of the motion and appearance features of the pyramid pooling layer. Finally, it used the decision fusion strategy to obtain the final motion recognition result. Experiments were performed on the UCFlOl and HMDB51 datasets, it achieved 94. 2% and 70. 5% recognition accuracy, respectively. The experimental results show that the improved network model achieves high recognition accuracy in video based human motion recognition tasks. [ABSTRACT FROM AUTHOR]

Accurate and timely winter wheat yield estimation has significant effect to grain markets and policy. Most crop estimation methods can be divided into two categories, one is based on the crop model and the other is the statistical learning method. For statistical learning methods with recent advances in deep learning, convolutional neural network (CNN) have become state-of-the-art algorithms. can extract the depth-dependent features of crop growth. However, the pivotal challenge is to combine remote sensing images with CNN. In this paper, we employ the method of histogram dimensionality reduction and time series fusion to generate the input layer. The experiment firstly performed projection transformation, splicing, mask, fusion, and clipping for 6 different MODIS images in the research area from 2006 to 2016, and then generated 21 600 fusion images of 12 bands (surface reflectance data of 7 different wavelengths in MOD09A1, surface temperature of day and night in MYD11A2, NDVI and EVI in MOD13A1, and FPAR in MOD15A2H). Then, the sensitivity range of winter wheat growth in each band is divided into 36 sections, and the histogram statistics are used to reduce the dimension to generate a vector of length 36, so the remote sensing image generates a matrix of 36×36×12 in the 228-day growing season. The corresponding time and regional statistics are applied as the output layer to construct a complete sample. The yield estimation sample database of 12 indices in the winter wheat region of north China (60 prefecture-level cities) from 2006 to 2016 was constructed, and the training set and verification set were divided into 10:1 for the training and evaluation of yield estimation model. Finally, the neural network structure is designed according to the sample, which consists of the input layer, 7 convolution layers (c1-c7), 7 activation layers, 7 batch normalization layers, 3 dropout layers, 2 full connection layers, and output layer. The number of c1-c7 convolution kernels is 64, 64, 128, 128, 256, 256, 256, the convolution kernel size is 3×3 dpi, and the sliding step length is 2, 1, 2, 2, 2, 1 and 2 respectively, 1 zero paddings per convolutional layer. At the same time, batch normalization and Relu function activation are performed on each convolutional layer, and the Dropout layer is used in the fully connected layer. The results show that: 1) The root-mean-square error (RMSE) and coefficient of determination (R²) of the convolutional neural network model on the training set are 183.82 kg/hm² and 0.98 respectively. In the validation set, RMSE and R² are 689.72 kg/hm² and 0.71. 2) With the same neural network structure, the average RMSE of the estimated samples from 2006 to 2016 trained as validation sets for 11 models was 772.03 kg/hm². The error of the yield estimation model was the largest in 2007 and the smallest in 2012, and the RMSE was 920.45 kg/hm² and 632.08 kg/hm² respectively. Crop estimation algorithm based on CNN has high robustness and precision; 3) The accuracy analysis of prediction yield at the municipal level of different provinces in three temporal points of 2007, 2012 and 2016 indicates that the model has higher accuracy in most areas of the northern winter wheat region, especially, RMSE of Hebei and Shandong provinces is approximately 500 kg/hm². The result shows that CNN is well applied to the estimation of winter wheat production. This is a great thought of remote sensing combined with the deep learning algorithm. This method can be used to estimate yield by remote sensing in different scales and regions. Compared with the traditional method, this “start-to-end” learning method has the advantage of synergy and can obtain the optimal estimation model relative to the whole area. Meanwhile, As data accumulates, the estimation accuracy will be continuously improved, and it has a good application prospect in the national agricultural production forecast. [ABSTRACT FROM AUTHOR]

Accurate and timely winter wheat yield estimation has significant effect to grain markets and policy. Most crop estimation methods can be divided into two categories, one is based on the crop model and the other is the statistical learning method. For statistical learning methods with recent advances in deep learning, convolutional neural network (CNN) have become state-of-the-art algorithms. CNN can extract the depth-dependent features of crop growth. However, the pivotal challenge is to combine remote sensing images with CNN. In this paper, we employ the method of histogram dimensionality reduction and time series fusion to generate the input layer. The experiment firstly performed projection transformation, splicing, mask, fusion, and clipping for 6 different MODIS images in the research area from 2006 to 2016, and then generated 21 600 fusion images of 12 bands (surface reflectance data of 7 different wavelengths in MOD09A1, surface temperature of day and night in MYD11A2, NDVI and EVI in MOD13A1, and FPAR in MOD15A2H). Then, the sensitivity range of winter wheat growth in each band is divided into 36 sections, and the histogram statistics are used to reduce the dimension to generate a vector of length 36, so the remote sensing image generates a Matrix of 36×36×12 in the 228-day growing season. The corresponding time and regional statistics are applied as the output layer to construct a complete sample. The yield estimation sample database of 12 indices in the winter wheat region of north China (60 prefecture-level cities) from 2006 to 2016 was constructed, and the training set and verification set were divided into 10:1 for the training and evaluation of yield estimation model. Finally, the neural network structure is designed according to the sample, which consists of the input layer, 7 convolution layers (c1-c7), 7 activation layers, 7 batch normalization layers, 3 dropout layers, 2 full connection layers, and output layer. The number of c1-c7 convolution kernels is 64, 64, 128, 128, 256, 256, 256, the convolution kernel size is 3×3 dpi, and the sliding step length is 2, 1, 2, 2, 2, 1 and 2 respectively, 1 zero paddings per convolutional layer. At the same time, batch normalization and Relu function activation are performed on each convolutional layer, and the Dropout layer is used in the fully connected layer. The results show that: 1) The root-mean-square error (RMSE) and coefficient of determination (R2) of the convolutional neural network model on the training set are 183.82 kg/hm2 and 0.98 respectively. In the validation set, RMSE and R2 are 689.72 kg/hm2 and 0.71. 2) With the same neural network structure, the average RMSE of the estimated samples from 2006 to 2016 trained as validation sets for 11 models was 772.03 kg/hm2. The error of the yield estimation model was the largest in 2007 and the smallest in 2012, and the RMSE was 920.45 kg/hm2 and 632.08 kg/hm2 respectively. Crop estimation algorithm based on CNN has high robustness and precision; 3) The accuracy analysis of prediction yield at the municipal level of different provinces in three temporal points of 2007, 2012 and 2016 indicates that the model has higher accuracy in most areas of the northern winter wheat region, especially, RMSE of Hebei and Shandong provinces is approximately 500 kg/hm2. The result shows that CNN is well applied to the estimation of winter wheat production. This is a great thought of remote sensing combined with the deep learning algorithm. This method can be used to estimate yield by remote sensing in different scales and regions. Compared with the traditional method, this “start-to-end” learning method has the advantage of synergy and can obtain the optimal estimation model relative to the whole area. Meanwhile, As data accumulates, the estimation accuracy will be continuously improved, and it has a good application prospect in the national agricultural production forecast. [ABSTRACT FROM AUTHOR]

This paper tried to propose a method to solve the problem of sentiment classification by integrating word embedding and convolutional neural network (CNN). First of all, the method accomplished a training process with skip-gram model to generate word embedding of each word in the dataset. Then, it created a two-dimensional feature matrix which was the combination of word embedding of each word in a training sample as the input of CNN model. Each iteration process of training, entries of feature matrix would also update as part of model parameters. Secondly, this paper proposed a CNN structure which was mainly composed of three different sizes of convolution kernels so as to complete the automatic extraction process of a variety of local abstract features. Compared with traditional machine learning algorithms, the proposed word embedding and CNN based sentiment classification model has successfully improved classification accuracy by 5.04%. [ABSTRACT FROM AUTHOR]

To improve the accuracy of the pedestrian detection method. this paper proposed a rectangular input of convolution neural network enhance the new pedestrian detection method based on the depth residual network and YOLO object detection method. The rectangular input helped the model gain the pedestrian characteristics expression by analyzing the expression and distribution characteristics of pedestrians in the images. The depth residual network with pre-activation for YOLO object detection improved the feature extraction ability through more layers of convolution neural networks. Hybrid dataset training and cluster anchor boxes could also improve the pedestrian detection performance. Hie test results of INRIA dataset prove that the method has better detection performance than the popular pedestrian detection methods, the index of false positive per image can reduce to 13.86% .improving ranging from 1.51% to 58. 62% in varying degrees. [ABSTRACT FROM AUTHOR]

In order to further improve the performance of video image super-resolution reconstruction and study the reconstruction of spatial resolution of video images by using the characteristics of convolution neural network .this paper proposed a video image reconstruction model based on convolution neural network. Hie model adopted the pre-training strategy to initialize the parameters. And it carried out the training processing both on the spatial and temporal dimensions of the multi-frame video images at the same time. It extracted the characteristics of the main motion information, learnt and made full use of the information inter the frames for improved performance. And it used the adaptive motion compensation algorithm to optimize the output of the channel to obtain the reconstructed center frame image with high resolution. The experimental results show that the average of objective evaluation indexes for video image reconstruction improves with a rather clear margin (PSNR + 0. 4 dB/SSIM + 0.02) ,and the edge of the fuzzy phenomenon in video reconstruction image for the subjective visual effect is effectively reduced. Compared with other traditional algorithms,it both obviously improved the evaluation of the objective indexes and subjective visual effect of the reconstructed image. Providing a novel architecture based on convolution neural network for video image super-resolution, which provides an exploration for the further study of video image super-resolution reconstruction based on the deep learning method. [ABSTRACT FROM AUTHOR]

Aiming at the problem of complex feature extraction and low accuracy in human action recognition, this paper proposed a network structure combining batch normalization algorithm with GoogLeNet network model. Applying batch normalization idea in the field of image classification to action recognition field, it improved the algorithm by normalizing the network input training sample by mini-batch. For convolutional network, RGB image was the spatial input, and stacked optical flows was the temporal input. Then, it fused the spatio-temporal networks to get the final action recognition result. It trained and evaluated the architecture on the standard video actions benchmarks of UCF101 and HMDB51, which achieved the accuracy of 93. 50%and 68. 32%. The results show that the improved convolutional neural network has a significant improvement in improving the recognition rate and has obvious advantages in action recognition. [ABSTRACT FROM AUTHOR]

Based on convolutional neural networks， this paper proposes innovative ideas for diagnosing intracranial tissue and organ diseases. The images of Parkinson’s disease patients were preprocessed by gray mapping， histogram equalization， improved wavelet denoising and image enhancement. A convolutional neural network is designed and built on the basis of VGG-Net network model. Regularization strategy is adopted to avoid over-fitting problem. Then the ten-fold cross-validation of the network model is carried out five times with MRI image set of patients. And feature visualization is performed by deconvolution network to dig out latent features of diseases. The result and objective evaluation show that the network can make a good diagnosis of MRI image for Parkinson’s disease. [ABSTRACT FROM AUTHOR]

In order to overcome the shortcomings that low detection accuracy and slow network training of the real-time object detection algorithm YOLO, this paper proposed an algorithm that added the batch re-normalization to the YOLO network structure. It combined the characteristic of batch re-normalization, which had the strength to deal with small or non-i. i. d. minibatches. The improved algorithm treated the feature maps, which generated from the convolutional layers, as activations, and then batched renormalize the activations, meanwhile it removed the dropout from the original network structure and increased learning rate. The experimental results show that the proposal algorithm has better detection accuracy and faster than before, and furthermore, it can decrease the model training time and the requirement of hardware equipment. [ABSTRACT FROM AUTHOR]

China is the largest country for cultivating kiwifruit, and Shaanxi Province provides the largest production, which accounts for approximately 70% of the production in China and 33% of the global production. Harvesting kiwifruit in this region relies mainly on manual picking which is labor-intensive. Therefore, the introduction of robotic harvesting is highly desirable and suitable. The fast and effective recognition of kiwifruit in the field under natural scenes is one of the key technologies for robotic harvesting. Recently, the study on kiwifruit recognition has been limited to a single cluster and multi clusters in the field have seldom been considered. In this paper, according to growth characteristics of kiwifruit grown on sturdy support structures, an RGB (red, green, blue) camera was placed around 100 cm underneath the canopy so that kiwifruit clusters could be included in the images. We proposed a kiwifruit image recognition system based on the convolutional neural network (CNN), which has a good robustness avoiding the subjectivity and limitation of the features selection by artificial means. The CNN could be trained end to end, from raw pixels to ultimate categories, and we optimized the critical structure parameters and the training strategy. Ultimately, the network was made up of 1 input layer, 3 convolutional layers, 2 sub-sampling layers, 1 full convolutional layer, and 1 output layer. The CNN architecture was optimized by using batch normalization (BN) method, which normalized the data distribution of the middle layer and the output data, accelerating the training convergence and reducing the training time. Therefore, the BN layers were added after the 1, 3 and 5th convolutional layer (Conv1, Conv3, and Conv5 layer) of the original LeNet network. The size of all convolutional kernels was 5×5, and that of all the sub-sampling layers was 2×2. The feature map numbers of Conv1, Conv3, and Conv5 were 6, 16 and 120, respectively. After manual selection and normalizing, the RGB image of kiwifruit was transferred into a matrix with the size of 32×32 as the input of the network, stochastic gradient descent was used to train our models with mini-batch size of 100 examples, and momentum was set as 0.9. In addition, the CNN took advantages of the part connections, the weight sharing and Max pooling techniques to lower complexity and improve the training performance of the model simultaneously. The network used rectified linear units (ReLU) as activation function, which could greatly accelerate network convergence. The proposed model for training kiwifruit was represented as 32×32-6C-2S-16C-2S-120C-2. Finally, 100 images of kiwifruit in the field (including 5918 fruits) were used to test the model, and the results showed that the recognition ratios of occluded fruit, overlapped fruit, adjacent fruit and separated fruit were 78.97%, 83.11%, 91.01% and 94.78%, respectively. The overall recognition rate of the model reached 89.29%, and it only took 0.27 s in average to recognize a fruit. There was no overlap between the testing samples and the training samples, which indicated that the network had a high generalization performance, and the testing images were captured from 9 a.m. to 5 p.m., which indicated the network had a good robustness to lightness variations. However, some fruits were wrongly detected and undetected, which included the fruits occluded by branches or leaves, overlapped to each other and the ones under extremely strong sunlight. Particularly, 2 or more fruits overlapped were recognized as one fruit, which was the main reason to the success rate not very high. This phenomenon demands a further research. By comparing with the conventional methods, it suggested that the method proposed obtained a higher recognition rate and better speed, and especially it could simultaneously identify multi-cluster kiwifruit in the field, which provided significant support for multi-arm operation of harvesting robotic. It proves that the CNN has a great potential for recognition of fruits in the field. [ABSTRACT FROM AUTHOR]

Deep learning method has very excellent ability of image feature extraction. In order to get rid of disadvantages of traditional methods require a priori knowledge, this paper proposed an image feature extraction algorithm based on the fusion AutoEncoder and convolutional neural network (CNN ). The method introduced a fast sparsity control technique to AutoEncod-er and utilized AutoEncoder to train the basic elements of image and initialized the convolution kernel of CNN. Meanwhile, the algorithm added filtering mechanism to the CNN to keep the sparsity of output characteristics. The results of experiments point out that this method achieves good performance on the Minist handwritten digital library and the Yale face database. Furthermore, the advanced experimental outcomes indicate that the feature extraction model included the filtering technique is more effective than the model without filtering mechanism by using cross-validation with T test. [ABSTRACT FROM AUTHOR]

To solve the problem that the loss of image feature information and improve the accuracy of image retrieval, when the convolutional neural network (CNN) was used to extract the feature information of the image, this paper proposed an image retrieval algorithm based on improved convolutional neural network LeNet-L. First, it improved LeNet-5 convolution neural network structure, increased depth of network structure. Then, it pre-trained the deep convolutional neural network LeNet-L to extract the high-level semantic features. At last, it obtained the similar images by distance function between the image being retrieved and the one in image database. In Corel dataset, compared with the original model method and the traditional image retrieval method based on SVM and active learning, the proposed method had a higher accuracy. The experimental results show that the improved convolutional neural network has a better retrieval effect. [ABSTRACT FROM AUTHOR]

Plant leaf diseases are a serious problem in agricultural production. To solve this problem and prevent diseases deterioration, accurate identification of diseases types is of great significance. In this paper, we proposed a recognition model of plant leaf diseases based on convolutional neural network (CNN), which combines the batch normalization and global pooling methods. The parameters of the traditional CNN model are large and have difficulty to converge. The proposed model was modified in the traditional structure of the CNN, which could optimize the training time and achieve the higher accuracy, and also reduce the size of model. In order to speed up the training convergence, we used the batch normalization layers. We put the input of every convolutional layer in batch, calculated the mean and variance of the batch, and then normalized this batch. We reduced some feature maps of some layers and removed the last full connect layer, with the global pooling layer instead. The proposed model has 5 convolutional layers and 4 pooling layers. In the last pooling layer pool5, the same kernel size of convolutional layer Conv5 was used to take advantage of the information of Conv5's feature map comprehensively. For the image preprocessing, we had zoomed, flipped and rotated the original pictures of dataset randomly to get the augmented dataset, and used the 80% of pictures as the train dataset and the rest as the test dataset. These pictures were quantized to 256×256 dpi for CNN training, and the original dataset and augmented dataset were used to train models. To look for the best size of the first layer kernel, in the first convolutional layer, different kernel sizes i.e. 11×11, 9×9 and 7×7 dpi were used respectively. Furthermore, we chose the type of global pooling layer, like max pooling and average pooling. Then we designed 8 models with different Conv1 kernel sizes or global pooling types. To further improve the efficiency of this model, besides using the Gaussian initialization, we used the other common type of convolutional initialization such as Xavier initialization, and also used the PRelu activation function for each convolution layer. So the optimal model could be selected to recognize the 26 kinds of leaf diseases which involved 14 kinds of plants, and then we analyzed the model's convergence rate, memory usage and robustness. After the experiment, we compared the test accuracy between the traditional model and the proposed model based on original dataset and augmented dataset. The proposed model could accelerate the training convergence, and the test accuracy could achieve about 90% while the traditional model was only about 77% after 3 training epochs. Different kernel sizes of Conv1 had little impact on the accuracy but small kernel was proved to be more beneficial to the recognition of plant diseases, which could get more texture features than the big kernel size filter, and average pooling also made better results than max pooling. We got the best performance model which used the 9×9 dpi kernel size and global average pooling layer. To show the proposed model's performance, we tested the accuracy on each class, and the mean accuracy of augmented test dataset was 99.56%, and the weighted average score of recall and precision rate achieved 99.41%. The proposed model had the size of only 2.6 MB. In addition, compared with the traditional methods, the change of the spatial position of the pictures had little effect on the performance of the improved model, and the proposed model could identify different diseases of various plant leaves. The results show that the model has higher recognition accuracy and stronger robustness, and can be used for the identification of plant leaf diseases. [ABSTRACT FROM AUTHOR]

Tea is one of important cash crops in China. Computer vision plays an important role in pest detection. Automatic classification of insect species in field is more difficult than the generic object classification because of complex background in filed and high appearance similarity among insect species. In this paper, we proposed an insect recognition system on the basis of image saliency analysis and a deep learning model, i.e. convolutional neural network (CNN), which has a good robustness with avoiding the features selected by artificial means. In image saliency analysis, we segmented the original images into super-pixel regions firstly. Then we quantized each RGB (red, green, blue) color channel and made them have 10 different values, which reduced the number of colors to 1000, and sped up the process of the color contrast of the pest objects and the background at region level. Finally, we obtained the saliency value of each region by combining their color contrast and spatial distances. The saliency values of all regions in each image were used to construct a saliency map, which was offered as the initial area for GrabCut algorithm to define the segmentation result and localize the pest object. The images after localization were quantized to 256×256 dpi for CNN training and classifying. CNN was trained end to end, from raw pixels to ultimate categories, thereby alleviating the requirement to manually design a suitable feature extractor. Based on theoretical analysis and experimental evaluation, we optimized the critical structure parameters and training strategy of CNN to seek the best configuration. The overall architecture included a number of sensitive parameters and optimization strategies that could be changed. We determined the local receptive field size, number, and convolutional stride as 7×7 dpi, 64 and 4, respectively. Dropout ratio for the fully-connected layers was 0.7. The loss function Softmax was fit for the pest classification system. To further improve the practical utility of CNN, we focused on structural changes of the overall architecture that enabled a faster running with small effects on the performance. We analyzed the performance and the corresponding runtime of our model by reducing its depth (number of layers) and width (number of convolution kernel in each layer). Removing the fully-connected layers (FC6, FC7) made only a slight difference to the overall architecture. These layers contained almost 90% of the parameters and when they were removed, the memory consumption decreased to 29.8 MB. But, removing the intermediate convolutional layers (Conv2, Conv3, Conv4, Conv5) resulted in a dramatic decrease in both accuracy and runtime. This suggested that the intermediate convolutional layers (Conv2, Conv3, Conv4, Conv5) constituted the main part of the computational resource, and their depth was important for achieving good results. We then investigated the effects of adjusting the sizes of all convolutional layers, and the filters in each convolutional layer were reduced to 64 each time. Surprisingly, all architectures showed significant decreases in running time with relatively small effects on performance. Finally, we determined the convolution kernel numbers of Conv2-5: 64-192-192-64. On the test set of tea field images, the architecture before and after shrinking respectively achieved the average accuracy (AA) of 0.915 and 0.881, respectively, superior to previous methods for pest image recognition. Further, after optimization the running time reduced to 0.7 ms and the memory required was 6 MB. [ABSTRACT FROM AUTHOR]

In ihe task of relation extraction, traditional statistical methods have difficulties in getting perfect features manually, while deep learning methods strongly depend on the representational capacity of single word vector. To solve the problems mentioned above, this paper proposed a novel model called multi-channel convolutional neural network (CNN). Firstly, the model used different word vectors to represent sentence as input for different channels. Secondly, it extracted features automatically through CNN. Finally, it obtained relation types by using softmax classifier. The proposed model could capture more semantic information thus it learned more distinctive features automatically comparing to other models. The experiment results on Sem E-val-2010 Task 8 datasets show that the proposed model is more suitable for relation extraction task than normal models which only utilize single word vector. [ABSTRACT FROM AUTHOR]

The classification accuracy of the traditional visual features can not meet the requirements in the application of modern military affairs due to the extremely high similarity of the different objects in the battlefield and low recognition rate in complex conditions. This paper presented a new architecture of the convolutional neural network (CNN) based on PCA whitening for solving the classification problem of large-scale images which contained some specific military objects. It efficiently eliminated the relativity of sample data. enhanced the learning ability and improved the accuracy of the object recognition. It tested and evaluated the new CNN model with the large-scale data from military images and compared with the traditional methods. The experiment results show that the algorithm has higher recognition rate on military object recognition. [ABSTRACT FROM AUTHOR]