Search

The level of face recognition technology continues to improve, and has achieved an ideal recognition rate in identity authentication, human-computer interaction and other industrial applications, and the market scale keeps growing. However, the occlusion problem in real scenes has not been completely solved. How to restrain or eliminate the negative impact of occlusion on key facial feature, is one of the hot spots in the field of face recognition. Aiming at the problem of lack of face structure information caused by occlusion, this paper reviews the occluded face recognition datasets and the occluded face recognition methods. Firstly, this paper introduces and analyzes some important new occluded face recognition datasets; Secondly, this paper summarizes the traditional learning methods and deep learning based methods to solve occlusion problems, and emphasizes the deep learning based robustness feature extraction and occluded facial information recovery. Finally, this paper summarizes and analyzes the advantages and disadvantages of relevant methods, points out the problems and challenges of occluded face recognition, and prospects the future research directions. [ABSTRACT FROM AUTHOR]

In order to address the shortcomings of the facial landmarks detection models, which cannot model the relations between long-distance landmarks, this paper proposed a parallel multi-branch architecture combining with convolution and Transformer for facial landmarks tasks, called MCTN, it utilized the dynamic attention mechanism to model the long-distance relations between facial landmarks. The multi-branch parallel structure designing allowed MCTN to include shared weights, global information fusion and other merits. What’s more, this paper proposed the novel Transformer structure, Deformer, which could make the MCTN focused attention weights faster on sparse and meaningful locations and solved the problem of slow convergence of Transformer. MCTN reached 4.33%,3.12% and 3.15% normalized average error respectively on the WFLW,300W and COFW datasets, the results show that MCTN utilizes Transformer with CNN multi-branch parallel structure and Deformer structure to dramatically outperform other facial landmarks localization algorithms based on convolution network. [ABSTRACT FROM AUTHOR]

Local feature information plays an important role in image segmentation, however, the referring image segmentation task is dependent on the text expression, so it is impossible to extract local feature information directly from the original reference image. In order to solve this problem, this paper proposes a specific noun-guided local feature extraction deep neural network model（NgLFNet). The NgLFNet model can automatically mine the local feature information of the object to be segmented according to the key nouns in the input text expression. Specifically, the model first obtains key nouns in text through sentence analysis; secondly, extracting corresponding features through text and image encoders, and using the key nouns to obtain local features of high-interest regions through the multihead attention mechanism; then the multi-modal features are gradually fused to learn; Finally, the decoding and correction module uses the obtained local features to perform more detailed corrections on the prediction mask to obtain the final result. The method in this paper is compared with a variety of mainstream referring segmentation methods and the experimental results show that the proposed method improves the accuracy of text-based instance segmentation task. [ABSTRACT FROM AUTHOR]

There is a huge demand for weather cloud images in the observatory' s weather monitoring system. In order to solve the problems of model instability and loss of image features when the conventional generative adversarial network expands the dataset of the weather cloud images, this paper proposed a double-layer embedded adversarial image generation method based on SAU-Net DCGAN. This method consisted of two layers of networks which were nested within each other. Firstly,by the first layer of embedded network, it added the U-shaped network to the generator of the generative adversarial network. This network acted as the basic architecture and enhanced the feature recovery capability of the image by using the jump connection between the encoder and the decoder. Secondly,by the second layer of embedded network, it added SA to the U-shaped network. This attention mechanism reduced the complexity of the model by simplifying the parameters, improved effectively the feature loss in the dark part of the image. Finally, it developed a new weight calculation method to strengthen the connection between each features and improved the extraction of detail texture features from the images. The experimental results show that the quality of the images generated by this method is better than that of the conventional generative adversarial network in terms of sharpness and saturation. The evaluation indicators PSNR and SSIM have increased by 27. 06 dB and 0. 606 5 respectively. [ABSTRACT FROM AUTHOR]

For image denoising, this paper proposed CEANet which had a dual-channel dilated convolution with attention mechanism to solve the problem of information loss caused by deep neural network. Reserving block merged output feature maps of each layer to make up the loss of detail information during convolution. Dilated convolution achieved better balance between denoising performance and efficiency,extracting more features with less parameters and enhancing the representation capability of the model for noisy images. The sparse module of dilated convolution expanded the receptive field to extract significant structural information and edge features and recover details of complicated noisy images. The feature enhancement module based on attention mechanism further guided network for image denoising by fusing global features with local features. The experimental results show that CEANet achieves high peak signal-to-noise ratio and structure similarity mean value at Gaussian white noise level of 25 and 50, which can capture image detail information more efficiently and has better performance in edge retention and noise suppression. Through the above comparative experiments proved the effectiveness of the algorithm framework. [ABSTRACT FROM AUTHOR]

High-quality learning low-dimensional representation of nodes in the graph is a current research hotspot. The existing shallow model methods cannot capture the nonlinear relationship of the graph structure, and the graph convolution model in the graph neural network technology will cause an over-smoothing problem. At the same time, how to determine the role of different hop number relationships in graph representation learning is also a problem that needs to be solved in the research. To solve the above problems, this paper proposes a deep learning model based on T（T＞1） feedforward neural networks. The framework uses deep learning models to extract the nonlinear relationship of the graph structure, and T sub-models effectively capture the local and global（higher-order） relationship information of the graph, and they give different roles in the final vector representation to take advantage of different hop relations. Experimental results on vertex classification and link prediction tasks show that the framework is competitive with existing methods, the benchmark algorithm can be improved by about 20%. [ABSTRACT FROM AUTHOR]

As a key problem and hub link between problem discovery and problem solving, the current research on root cause analysis mainly includes two types of methods: data-driven and causality-driven. In view of the advantages of data-driven methods in narrowing the range of root causes, at present, root cause research mainly focusses on datadriven methods such as association rule mining, heuristic search, machine learning and deep learning. There are few root cause analysis from the perspective of causal knowledge, nor has it been based on method dimension. The inductive analysis of the root cause is carried out, and there is a lack of relevant research results. Therefore, this paper sorts out and summarizes the main results of root cause analysis in recent years, analyses the differences and advantages of root cause analysis under different method dimensions, and proposes a root cause analysis method that integrates causal knowledge, and combines asymmetric Shapley value with causal chain diagram to improve the accuracy of root cause analysis. Finally, the existing some research difficulties and development trends put forward scientific and meaningful future research directions. [ABSTRACT FROM AUTHOR]

In order to improve the search ability of AGV in complex unknown environment, this paper proposed an improved DDPG. It constructed the empirical playback matrix and double-layer network structure to improve the convergence speed of the algorithm ε-greedy in the green search strategy, to solve the local optimization problem of AGV in selecting the optimal action. To solve the problem of slow training speed of deep neural network, it applied priority sampling to depth deterministic strategy gradient algorithm. In order to solve the problem of high complexity of ordinary priority sampling, this paper proposed a small batch priority sampling method to train the network. In order to verify the effectiveness of the method, it used the grid method to model and compare the simulation experiments in different complex environments. It compared the loss function, iteration times and return value of different algorithms. The experimental results show that compared with the original algorithm, the improved algorithm reduces the loss function, reduces the number of iterations and increases the return value, which verifies the effectiveness of the algorithm. At the same time, it provides a new idea for AGV to complete the planning task more safely and quickly in complex environment. [ABSTRACT FROM AUTHOR]

Single-view 3D reconstruction based on deep learning is a research hot spot at present. In order to discover more high-frequency details, SDF-SRN algorithm introduces positional encoding, but neural network is easy to overfit without accurate supervision, and reconstructs uneven surface. To solve the problem, this paper proposed the network model based on sparse feature. The model enabled the network that preferred to overfitting to predict high-frequency residual by residual learning. The feature extraction network extracted sparse features and the global features. Then one hypernetwork took the sparse features as input and generated prediction shallow head. This shallow head predicted low-frequency part of signed distance function. Another hypernetwork took global features as input and generated another shallow head. This shallow head predicted high-frequency residual. It fused two predictions of shallow heads into final signed distance function. Spectrum analysis shows that the design purpose of network is achieved. Compared with other smooth surface reconstruction schemes, the network can achieve smoother surface reconstruction with enough details. It overcomes the overfitting of SDF-SRN. The qualitative and quantitative comparison with other advanced single-view reconstruction approaches show the superiority of the proposed approach. [ABSTRACT FROM AUTHOR]

The recommendation algorithm based on graph neural network can extract the association relationship between users and goods. Traditional methods can’t extract this relationship. At present, most of these algorithms ignore the general prefe-rences in the review data of users and products. In order to solve this problem, this paper proposed a new method. This me-thod used the graph neural network to extract association relations, and the advantage of deep learning to extract the general preferences, and carried out feature fusion to improve the recommendation effect. This paper conducted comparative experiments and ablation experiments on four sets of public data sets to verify the effectiveness of the proposed method. The evaluation indexes include the recall rate and normalized discounted cumulative gain. Experiments show that this method is more effective than the existing algorithms. The feature fusion of the two networks can improve the recommendation effect. [ABSTRACT FROM AUTHOR]

When the scale of data is large, the deep learning model will encounter the problem of time-consuming weight adjustment and easy to fall into the local optimal solution. In order to solve these problems, broad learning system comes into being. The broad learning system not only has the advantages of simple structure, fast training speed and high accuracy, but also has the advantage of incremental learning. This paper firstly introduced the background and development of broad learning system, then described the basic theory and implementation method of broad learning system, and compared its similarities and differences with deep network. Then this paper introduced the improvement of broad learning system on image classification, numerical regression, EEG signal processing, and analyzed the advantages and disadvantages of these algorithms. Finally, it summarized the shortcomings of existing broad learning algorithms, and prospected the future research direction. [ABSTRACT FROM AUTHOR]

This paper presented an image dehazing method based on deep learning to solve the problem that the remote sensing images had reduced image sharpness due to haze. Firstly, this algorithm distorted the original atmospheric scattering model to get an end-to-end defrosting model. Then, it unified several unknown parameters to one parameter and estimated the unknown parameters by using multiscale convolution neural network. Finally, it brought the parameter estimates into the defrosting model to get a haze-free image. For no reference image dataset, this paper used the existing dataset to preliminary train the network, then added the self-built dataset to secondary train the network. The experimental results show that, compared with the related defrosting algorithms, the proposed algorithm improves the visual effect and objective index to some extent, and effectively improves the clarity of remote sensing images in haze weather conditions. [ABSTRACT FROM AUTHOR]

This paper constructed a new TerraSAR-X dataset named GDUT-Nansha and then proposed a new lightweight semantic segmentation algorithm using synthetic aperture radar images（ SAR）. It is difficult to apply traditional deep learning models to SAR datasets with small volume due to the tremendous amount of parameters. A revision scheme is, therefore, put forward to deal with such problem. This paper accomplished an improved lightweight convolutional neural network model named revised weighted loss ENet（ RWL-ENet） for SAR images. The current study introduced a weighted loss function to solve the problem of imbalance of training datasets. Compared with other classical convolution neural network models, the efficiency and robustness of the new dataset were validated. Meanwhile, RWL-ENet model attained 0. 884,0. 804, and 0. 645 in terms of three quantitative metrics, including pixel accuracy（ PA）, mean pixel accuracy（ m PA）, and mean intersection over union（ m IoU）. In addition, the parameters of this new proposed model are much less than other classic network models. [ABSTRACT FROM AUTHOR]

Aiming at the traditional model parallel methods for accelerating convolution neural network (CNN) tend to have low utilization, this paper proposed PipeCNN, which accelerated CNN with software pipeline. Firstly, this paper studied the forward propagation and back propagation, and then explored data correlation during training. Secondly, it parallelized CNN with the support of software pipeline, and then analyzed two feasible gradient updating methods in PipeCNN. Finally, it used circular queue to realize communication between two layers and then proposed a task allocation algorithm to divide CNN into working parts. Experiments show that the method can obtain good speedup and utilization while ensuring the accuracy of the model. It shows that software pipeline can effectively solve the problem of low utilization in model parallel, and accelerate the training of CNN. [ABSTRACT FROM AUTHOR]

Due to the low randomness of character distribution and high randomness of word combination in word-based DGA domain names, the traditional detection methods are difficult to identify them. The method based on deep learning such as LSTM can capture domain's sequence features, but it lacks the word combination features and has low detection accuracy. In order to solve above problems, this paper proposed a new DGA domain name detection method based on fusion embedding layer. In domain embedding stage, it carried out a fusion vector representation based on characters and word segmentation technology . Then, it constructed a hybrid deep learning model based on CNN and BiGRU. The experimental results show that the proposed model has increased by 3 . 1 % and 4. 3% respectively in accuracy compared with the models based on CNN or LSTM alone. When dealing with the matsnu, suppobox and ngioweb DGAs, the proposed model is more effective. [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]

In order to solve the problem that the existing super-resolution algorithm based on deep learning didn't make full use of the feature information of each level, resulting in low reconstruction accuracy and large parameter quantity, this paper proposed a double dense connection structure named densely channel attention skip connection network. In the inner structure of the network, it improved the original dense cascade block to generate a channel separable dense cascade block. The outer structure adopted a densely residual connection and attention mechanism to fuse the features extracted by the dense block to achieve the goal that less convolution layer and higher precision effect. This paper tested the network models on several benchmark datasets . The results show the proposed model has higher accuracy and fewer parameters than the other models. [ABSTRACT FROM AUTHOR]

CNN (convolutional neural network) is widely used in image classification tasks. Most existing CNN-based classification models are trained as flat N-way classifiers. However, the difference among different image categories limits the capacity of the classifier. To solve the problem, this paper proposed CT-CNN (confusion tree convolutional neural network) model which combined the CT with CNN. CT-CNN first established a CT to identify the confused categories. Then it embedded the hierarchical structure of CT into CNN model, which leaded the CNN training procedure to pay more attention on strongly confused categories. Experiments on public datasets prove that CT-CNN can overcome the limitation of uneven distribution of classification difficulty between categories of large-scale datasets and achieve better performance on complex large-scale image classification tasks. [ABSTRACT FROM AUTHOR]

Existing deep learning based single image super-resolution (SISR) models usually improve the fitting ability of the model by increasing the number of network layers, but fail to fully extract and reuse features, leading low quality of reconstructed images. To solve this problem, this paper proposed an image super-resolution model based on feature fusion and attention mechanism. This model used residual in residual (RIR) structure in feature extraction module. The feature extraction module of the network consisted of several residual groups. Each residual group consisted of several residual blocks. This module implemented local feature fusion in each residual group and global feature fusion between each group. In addition, this model introduced coordinate attention module into each residual block and spatial attention module into each residual group. It verifies that the model is able to fully extract features and reuse features. The final experimental results show that the model is superior to the existing models in objective evaluation indexes and subjective visual effect. [ABSTRACT FROM AUTHOR]

Aiming at the problem of low person re-identification accuracy caused by similar appearance and posture of person in surveillance videos, this paper proposed a video-based person re-identification method based on a graph model, which effectively utilized the time sequence information in the video to realize the information interaction across frames and intra-frame regions. Specifically, it used the correlation information between the cross-frame block regions to establish the inherent relationship between the regional nodes, and iteratively updated the regional information through feature propagation. On the other hand, in the metric learning process, it proposed a weighted loss function strategy, which optimized the binary allocation method (that is, discarding or retaining the sample) in the previous mining strategy into a continuous score allocation method, which solved the problem that the available samples were not used efficiently. Finally, it evaluated the model on MARS and DukeMTMC-VideoReID datasets, and the experimental results confirm the effectiveness of the proposed method. [ABSTRACT FROM AUTHOR]

Occlusion in a real environment is one of the main obstacles to accurately analyze and recognize facial expressions. In recent years, researchers have used deep learning technology to solve the problem of high misrecognition rate of facial ex-pressions under occlusion conditions. It mainly summarized the deep learning algorithm of occlusion facial expression recognition and occlusion-related issues. Firstly, this paper summarized the development status of facial expression recognition under partial occlusion conditions, expression representation, and the data set used to study occlusion expression. Secondly, it re-viewed the occlusion expression identify the latest developments in deep learning methods and analyzed the impact of occlusion on expressions. Finally, it summarized the main technical challenges, research difficulties and possible coping strategies. The purpose is to provide a more useful reference and benchmark for future research on occlusion expression recognition. [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]

In sound event detection tasks, the target event was susceptible to background noise, and wasn't present in a significantly high portion of sound signal flow. To solve the problem, this paper proposed a new method of sound event detection based on one-dimensional dense convolutional network (DenseNet ) with multi-level attention mechanism. Firstly, it used the one-dimensional DenseNet for frame-level detection, which was effective in finding the precise onset and offset time. Then, it introduced the multi-level attention mechanism in the one-dimensional DenseNet model, which made the attention-aware features from different modules change adaptively as layers went deeper. Therefore, the model could automatically select and attend on important frames for the targets while ignoring the unrelated parts (e. g. the background noise segments) . Finally, this paper evaluated the model using DCASE 2017 Task 2 development dataset. Results show that the overall performance of the proposed method has further improvement than the conventional deep learning method. [ABSTRACT FROM AUTHOR]

The single-frame image super-resolution（SISR） method based on deep networks is currently a hotspot in SR research. However, most of these methods mainly focus on exploring the network depth structure during feature extraction, ignoring the similarity between the feature layers in the intermediate space, and neglect the feature differences between the feature layers during reconstruction. This paper proposed a progressive network based on spatial feature transformation and back-projection reconstruction to solve this problem. The main feature of this method was to perform feature affine transformation on the feature space when extracting image features to increase the different similarities between feature layers, so as to obtain progressive features and spatial transformation features. In the image reconstruction stage, the reconstruction module adopted the strategy of multi-scale back projection to integrate the multi-source features of the image, which made the module pay more attention to the difference between features. Finally, the experimental results show that compared with most super-resolution methods, the proposed method has greatly improved the evaluation indexes such as PSNR/SSIM during image super-resolution reconstruction, and the texture information of the reconstructed image is more abundant. [ABSTRACT FROM AUTHOR]

Most of the current fake face detection technologies use deep learning to identify the feature differences between real videos and fake videos, and have achieved good results on uncompressed videos. But the detection performance on compressed video will be severely reduced. Therefore, this paper proposed a fake face video detection method based on improved triplet loss to tackle the above problem. Firstly, it used an artifact map generated by the artifact map generator to deepen the feature difference between the fake face and the real face. Secondly, it used improved triplet loss to solve the problem of hard samples that were difficult to detect correctly. Finally, it chose a deep learning network that was more suitable for fake face detection to extract convolutional features. Experiments on the FaceForensics++ dataset show that proposed method is better than other state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Image privacy leakage is an urgent problem to be solved in deep learning-based image retrieval systems. Using adversarial samples generated by the adversarial attack technology can achieve privacy protection to some extent. However, the existing targeted attack methods for image retrieval systems are susceptible to the quality and quantity of selected target samples, which can lead to poor attack effects. To solve this problem, this paper proposed a targeted adversarial attack method for image retrieval based on feature weighted aggregation. This method used the retrieval accuracy of the target image as weights to obtain the class features, which was generated by a small number of samples from the target class. Experimental results on two image retrieval datasets of RParis and ROxford show that the retrieval accuracy of the adversarial samples generated by this method improve by 38% on average compared with TMA method, and improve by 7.5% on average compared with DHTA method. [ABSTRACT FROM AUTHOR]

Aiming at the problem that the current image caption approaches can only describe the image with a black box architecture, which is difficult to control from the outside. This paper creatively transformed the problem of image caption into the problem of seq2 seq, and achieved the effect of controllable generation of image caption. It designed an entity set or entity sequence composed of image regions as the control signal, under the guidance of the block sentry with entity block switching and the adaptive attention mechanism with visual sentry, regularly input the control signal into the double-layer long short term memory (LSTM), it guided the model to generate the corresponding image caption in a controllable way. In addition, baseline used cross-entropy loss to train and early stop the model, and introduced the idea of reinforcement learning to solve the problem that the optimization target in training was inconsistent with the index in evaluating the algorithm effect, so as to further optimize the model effect. The experimental results show that: on MSCOCO and Flickr30k datasets, the proposed algorithm achieves great results in generating controllable image caption, caption quality and diversity. [ABSTRACT FROM AUTHOR]

Network representation learning aims to transform the nodes in the network into a low-dimensional vector space while maintaining the inherent properties of the network. Most of the existing methods are aimed at normal networks, ignoring the particularity of attribute bipartite networks and the highly non-linear characteristics of the network. To solve the above problem, this paper proposed a deep attributed bipartite network embedding method incorporating with multilevel structure information. Specifically, the algorithm introduced an extended weight matrix to fuse the explicit and implicit structure of the bipartite network with attribute information. Then, it utilized a deep auto-encoder model to capture the highly non-linear characteristics of the network. To maintain the global network structure, a deep auto-encoder reconstructed the second-order proximity. Meanwhile, the deep auto-encoder used the first-order proximity of the nodes as supervisory information to maintain the local network structure. Finally, the algorithm performed joint optimization to get the final representation vectors of nodes. The model was executed on the four datasets of Yelp, Douban Book, Douban Movie and MovieLens. Compared with the latest benchmark method, the average values of F1@10, MAP@10,MRR@10 and NDGG@10 of this model have improved by 4.29%、5.63%、6.26%、4.21%. [ABSTRACT FROM AUTHOR]

In order to solve the problem that the recommendation model cannot mine the diversity of user interest and capture the order information between the sequence of user behavior, and the interaction occurs at the element level rather than between the feature vectors, etc., based on multiple attention mechanism and location information, this paper proposed the x DMAL FM. Firstly, it extracted the feature depth from different subspaces by using the multi-head attention mechanism, and then location information could capture the sequential relationship between user behavior sequences. Finally, it used three public datasets to conduct comparative experiments and the AUC index to evaluate the results. The experimental results show that the proposed algorithm has better recommendation performance than the xDeepFM model and show the effectiveness and feasibility of the proposed algorithm. [ABSTRACT FROM AUTHOR]

Capsule network emphasizes the encoding of image features' spatial relationship, while its feature extraction module has difficulty in handling complex scenarios. In order to improve the performance of CapsNet, this paper proposed a capsule network with a self-attention feature extraction module ( SA-CapsNet) to solve the above problem. At first, it improved CapsNet by reducing the dimension of the capsule and adding an intermediate layer. Then mapped the SA module to the feature extraction layer of the capsule network to increase feature extraction capability. The results worked on the MNIST, Fashion MNIST, and CIFARlO datasets show that the classification accuracy rates are 99. 67%, 92. 21 %, and 82. 51 %, respectively. Additionally, results demonstrate the effectiveness of this proposed network, and the overall performance has been dramatically improved compared with CapsNet [ABSTRACT FROM AUTHOR]

In the scene of document verification，the standard deep learning face recognition method has low accuracy and poor real-time performance on embedded devices. To solve these problems，this paper proposes a modified efficient convolutional neural network（CNN）called Lightnet and adopts the transfer learning method. Lightnet is an efficient CNN module composed of depthwise separable convolution，linear bottleneck structure and attention module. After introducing the loss function AM-Softmax with additive angle margin， the network model can effectively solve the problems of redundancy parameter and vast calculation for standard CNN in the foundation of ensuring the high accuracy of face recognition. The transfer learning method can enhance the scene-identity face matching performance by freezing all the convolution layer weights of the pre-trained model and fine-tuning training in the self-made scene-identity face matching dataset. The experimental results show that the designed efficient scene-identity face matching algorithm has achieved good results in terms of verification accuracy，parameters and verification speed，and has good robustness in life scenarios. [ABSTRACT FROM AUTHOR]

In order to solve the problem of the attacker uses the generative adversarial network (GAN) technology to restore the data in the training process and leaks the user' s private information, this paper proposed a differential privacy protection for Wasserstein generative adversarial network with gradient penalty (WGAN-GP) method. This method added accurately calculated Gaussian noise to the gradient during the deep learning training process, and used gradient penalty to perform gradient correction to achieve differential privacy protection. It used WGAN-GP to generate data that similar to the original data. Experimental results show that under the premise of ensure the availability of the data, the method can effectively protect the privacy information of the data and generate data with better quality. [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]

To solve the problem that the zine dust cover handling hole when unload the sherardizing reactor automatically, the single object detection model or method can't detect much well, this paper proposed a special anti-dust-occlusion object detection method with variety of feature model to aid the unloading. Firstly, the depthwise separabile convolution was a substitute for the traditional convolution in Tiny-YOLOv3. It also adjusted the loss function to adapt the new training. Secondly, it judged occlusion of target area by pixel statistics and estimated current frame with the trajectory features and shape features. Finally, the prediction results optimized the results of object detection network with fusion rules. The results show that the improved fusion model perform best with the severe and prolonged occlusion on handling hole, the AP and AIOU reach 99.38% and 88.14% respectively. Furthermore, it's a real-time method, achieving the dynamic detection of the handling hole with serious dust occlusion effectively. [ABSTRACT FROM AUTHOR]

Objectives: Road information extraction based on vehicle trajectory data is one of the hotspots and difficulties in the field of geographic information data acquisition. Traditional methods are faced with the problems of high accuracy of trajectory data source, complex road extraction algorithm model, and poor adaptability of different road extraction model parameters. In order to solve the above problems, a trajectoryroad conversion model based on conditional generative adversarial nets is proposed, and we called it trajectory‐ to‐road translation with conditional generative adversarial nets( TR‐CGAN).Methods: Firstly, the trajectory data and the corresponding road data are raster processed in the sample area to construct the trajectory‐ road sample image pairs, then the parameters of TR‐CGAN are learned based on the sample data as the prior knowledge. Through the continuous iteration of the one player game, the optimal generation result is gradually approached. Before that, according to the characteristics of vehicle trajectory data, this paper uses the control variable method and enumeration method to analyze the parameters of U ‐Net generator depth, discriminator receptive field size and objective function in the conversion model, so as to obtain the optimal structure of TR‐CGAN.Results: Using the taxi track data in the third ring road of Zhengzhou city, the experiment results show that this proposed method can find new roads more effectively. At the same time, the trained TR‐CGAN is compared with the raster road extraction method, and it is found that our method has stronger adaptability of trajectory data in both the sparse and dense areas of the trajectory, and the accuracy of generated road is higher.Conclusions: Our proposed method can realize road extraction based on trajectory data, and has better data adaptability and accuracy. In the further research, we can increase the type of sample data, so that the road extraction model can learn to generate more road types, such as ring road types. The model can be further optimized to extract two lane roads or even lane roads. [ABSTRACT FROM AUTHOR]

In order to solve the problem of low accuracy and residual error in small object detection on UAV aerial images, this paper proposed a new kind of multi-scale small target detection method based on enhanced lower feature. Basing on Faster RCNN ResNet-50-FPN model, the algorithm enhanced the lower feature by designing the structure of DetNet-59 feature extraction network and Flat-FPN feature fusion network, and applied soft-NMS to face the appearance of overlapping small objects. From simulation test on VOC2007 and VisDrone2019, the method is able to increase m AP by 11% compared to the base model when time consumption is no more than 2%, and it also performs better in terms of accuracy than current common algorithms. It was proved that the algorithm can effectively improve the detection accuracy of small targets while ensuring real-time performance. [ABSTRACT FROM AUTHOR]

In order to solve the problems of insufficient modeling of entity and relationship dependence and the difficulty of extracting multiple relationships involved in existing j oint extraction methods of entities and relations hips, this paper designed a joint extraction framework based on deep learning. Firstly, for the problem of insufficient dependency modeling, the framework extracted entity co-occurrence features from the pre-trained corpus, and modeled the potential semantic relationship between entities and the dependency relationship between entities and relationships. Secondly, it included a novel pointer labeling method. This labeling method could represent the relationship category through a pointer. Since any entity could be pointed by multiple pointers, it was possible to mark overlapping entities in a piece of text and extracted multiple entity-relation triplets result. Finally, in ord er to effectively use the rich semantics of words and the information dependent on pointers, it d es igned a tag-aware attention mechanism was necessary, which incorporated word information from the coding layer and related co-occurrence semantic information. Compared with the joint extraction method at the forefront of research, the proposed method achieved an increase in F1 value on the Baidu DuIE test set. The experimental results show that the pointer labeling method can solve the problem of entity overlap to a certain extent. [ABSTRACT FROM AUTHOR]

To solve the problem that most models do not make full use of candidate entity ranking and ignore the subject-relation dependency in simple question answering over knowledge graph, this paper proposed a method based on entity ranking and joint fact selection. It divided the whole process into three steps: pattern extraction, entity ranking and joint fact selection. Firstly, it extracted the natural language question by the BILSTM-CRF, divided it into two parts: mention and pattern. Then, it sorted the candidate entities by using the literal and semantic similarity of the subject and the mention to select candidate facts. Finally, in order to select the most correct subject-relation in the candidate fact pool, the joint fact selection model enhanced this process with multi-level encoding. Experiments show that the method can significantly improve the accuracy, recall rate on the simple questions dataset. The experimental results indicate the feasibility of the proposed method for simple question answering over knowledge graph. [ABSTRACT FROM AUTHOR]

At present, most of the current image style transfer methods are supervised learning, the training data need to appear in pairs. And when the background of the image needs to be processed, the existing methods are too cumbersome. In order to solve these problems, this paper proposed a new method of unsupervised image style transfer based on image mask. In the experiment, it adopted the architecture of the cycle-consistent generative adversarial network, and used the Inception-ResNet structure as the basic component to design a new generative model with the built-in image mask. At last,it automatically reconstructed the background of the image and the learned abstract features through unsupervised learning. Experiments show that the proposed method can effectively separate and reconstruct the background and the learned abstract features of the image, and solves the problem of regional interference in the feature learning process, and achieves considerable visual effects. [ABSTRACT FROM AUTHOR]

This paper proposed an improved residual network (ResNet) expression recognition algorithm. The algorithm used small convolution kernels and a deep network structure to solve the problem of accuracy reduction with the increase of depth by the residual module. The experiment overcame the shortcoming of insufficient data through transfer learning, which could effectively prevent overfitting. The network architecture used a linear SVM for classification. The experiment used the ImageNet database to pre-train network parameters to have an excellent ability to extract feature. According to transfer learning, the algorithm used the FER-2013 database and the expanded CK + database to fine-tune and train network parameters, and overcame the problem that shallow networks rely on manual features and deep networks were difficult to train. The results show the recognition rates is 91.333% and 95.775% on the CK + database and the GENKI-4K database,respectively. The classification accuracy of SVM in CK + database is about 1% higher than that of softmax. [ABSTRACT FROM AUTHOR]

Categorical features always exist in the dataset and coding them is a key issue for solving problems efficiently by machine learning models. One-hot coding is a wide accepted method to convert the features into feature values, and however it attracted sparse space and meaningless value after coding. To improve the coding performance, this paper designed a novel coding method based on conditional probability after divid ing the features into zones, which was called CZT coding. The CZT coding calculated the conditional probability of each feature and then divided the features into several zones and finally coded the features in each zone. It mathematically proved that compared with the state-of-the-art method- one-hot coding, CZT coding reduced the code length by at least the mean of feature spaces and the issue switched into an easier one after CZT coding for the following machine learning model. Finally, it used th e same neuron network as the classifier, the performance of CZT coding and one-hot coding was compared by using the Titanic dataset. The result shows that CZT coding makes the classifier performs better both on the accuracy and steadiness . [ABSTRACT FROM AUTHOR]

In order to solve the problem that the speed and accuracy of the current similarity video retrieval methods still cannot meet the requirements of users, this paper proposed a new similarity video quick retrieval method, which combined the three-dimen sional convolutional neural network with the hash learning method and applied to video data. It could not only quickly learn the video spatiotemporal feature representation, but also greatly shorten the video retrieval time. The experimental results on the set show that the similarity retrieval performance of the video using the proposed method is superior to the current mainstream methods. [ABSTRACT FROM AUTHOR]

In order to solve the problem of action classification prone to confusion after mapping the four-dimensional depth information to two-dimensional space, this paper proposed a method for human action recognition based on deep learning. Firstly, the method constructed spatially stru ctured dynamic depth images for dimension reduction. Then, it proposed the deep convolution neural network with joint cost function, which combined the cross entropy loss function and the central loss function as cost function, to gu ide the convolution layer to learn more discriminative deep features. The experimental results evaluated on the public MSRDailyActivity3D and SYSU 3D HOI dataset show that the method obtains a better performance compared with oth er existing methods, which validates the effectiveness and robustness of the method. The method effectively solves the problem of action classification prone to confusion. [ABSTRACT FROM AUTHOR]