Search

This paper proposed a self-supervised training method based on inter-frame consistency to solve the problem that the current inverse rendering supervised learning method is challenging to obtain labels and has poor generalization ability. Due to the ill-posed nature of the inverse rendering problem, this paper introduced additional albedo consistency loss and cross-rendering loss to strengthen the self-supervised network, the main idea of which was to enforce inter-frame consistency constraints on image sequences with continuous illumination changes. The method performed image projection and warping between adjacent frames through pose maps and depth maps between image frames. This method established constraints between adjacent frames and used Siamese training to ensure photometric invariance consensus estimate. This paper used a fully convolutional neural network to recover geometry, reflectivity, and illumination from indoor video sequences. The method trained the self-supervised network using a collection of unlabeled consecutive frame images and incorporating a differentiable renderer, making the network learn in a self-supervised manner. Compared with other mainstream methods, quantitative and qualitative experimental results show that the proposed method performs better on multiple benchmarks. [ABSTRACT FROM AUTHOR]

Generative adversarial network (GAN), as an emerging generative model, has been gradually developed and applied in the fields of image generation, 3D reconstruction, cross-modal conversion, etc. It effectively solves the problem of inefficiency of conventional convolutional neural networks in image-generating tasks and fills the shortage of deep learning in the field of image generation. In order to help subsequent researchers quickly and comprehensively understand GAN, this paper sorted out the improved model of GAN based on the literature in recent years. It firstly introduced the basic principles of GAN from two perspectives of network structure and objective function, then elaborated and summarized various derivative models of GAN from two major perspectives of improvement and application types. Secondly it summarized and analyzed the quality and diversity of generated images from the perspectives of subjective qualitative, objective quantitative and task-specific evaluation. Finally, this paper discussed some core issues and latest research progress of GAN series models in recent years and analyzed the future development trend. [ABSTRACT FROM AUTHOR]

Aiming at the current hand pose estimation which doesn’t fully use 2D information to assistance 3D hand mesh reconstruction, this paper further introduced HOG feature of the RGB image based on the use of hand Mask and 2D heatmaps. This paper designed the framework of multi task learning CNN to extract the 2D information, and used the topology of the hand to fused the relevant information. In order to solve two problems: the consistency of hidden Embeddings dimensions in Transformer Encoder as well as requires heavy computational resources, the model introduced a new MLP (Multi-Layer Perception) module into transformer encoder to increase the length of Embeddings and decrease the dimensions of Embeddings, thereby the hand Mesh was completed in the way from coarse to fine. The new method proposed in this paper has achieves better results in two Frei-HAND and RHD datasets. [ABSTRACT FROM AUTHOR]

In order to solve the problem of strong popularity bias in recommendation lists generated by Bayesian personalized ranking algorithm under the condition of unbalanced data, this paper designed a hybrid recommendation algorithm based on feature embedding to remove popularity bias. Firstly, this paper used convolutional neural network to extract user and item features, determine user preferences, and fill original unbalanced data according to user preferences. Secondly, this paper embedded the user preference features extracted from convolutional neural network into the Bayesian personalized ranking algorithm for hybrid recommendation. Finally, it trained the mixed recommendation model with score filled data, and obtained the personalized ranking list without popularity bias. In order to verify the performance of the algorithm, it conducted the analysis and comparison experiments on Movielens-100 K and Movielens-1 M. And the experimental results show that the popularity bias is reduced by about 50%～60% and the accuracy is improved by about twice. [ABSTRACT FROM AUTHOR]

The U-turn task is one of the contents of autonomous driving research, and most of the solutions under the standard roads in cities cannot be implemented on non-standard roads. Aiming at solving this problem, this paper establishes a vehicle U-turn dynamical model and designs a multi-scale convolutional neural network to extract feature maps as the input of the agent. In addition, for the sparse reward problem in the U-turn task, this paper proposes a hierarchical proximal policy optimization algorithm that combines hierarchical reinforcement learning and proximal policy optimization algorithm. In experiments with simple and complex scenarios, this algorithm learns policies faster and has a higher success rate of U-turn compared to other algorithms. [ABSTRACT FROM AUTHOR]

Most existing text classification algorithms based on graph neural network ignore the problem of over-smoothing, and ignore the problem of information loss due to graph topology, resulting in poor classification performance. To solve this problem, this paper proposed a method to measure the smoothness of multiple text graph representations WACD and a regularization term RWACD to suppress over-smoothing. Subsequently, this paper proposed an attention and residual-based network structure ARS to compensate for the loss of textual information due to graph topology differences. Finally, this paper proposed a graph convolutional neural network text classification algorithm RA-GCN. RA-GCN used ARS to fuse text representations in the graph representation learning layer, and used RWACD in the readout layer to suppress over-smoothing. This paper conducted experiments on 6 Chinese and English datasets. The experimental results demonstrate the classification performance of RA-GCN, and the effects of RWACD and ARS are verified through multiple comparative experiments. [ABSTRACT FROM AUTHOR]

In image retrieval based on deep hashing, to solve the problems of low feature extraction efficiency in convolutional neural networks (CNN ) and underutilization of feature correlation, this paper proposed a novel method combining sparse difference network and multi-supervised hashing( SDNMSH ), and used it for efficient image retrieval. SDNMSH took pairs of images as training inputs, and guided hash codes learning through an elaborately designed sparse difference convolutional neural network and a supervised hash function. The sparse difference convolutional layer and the vanilla convolutional layer composed the sparse difference convolutional neural network. The sparse difference convolutional layer could quickly extract rich feature information, to achieve efficient feature extraction of the entire network. At the same time, in order to make full use of the pairwise correlation of semantic information and features, so as to promote the feature information extracted by the network to be more effectively transformed into discriminative hash codes, and then to achieve efficient image retrieval by using SDNMSH, this paper adopted a multi-supervised hash( MSH) function and designed an objective function for this purpose. Extensive experimental results on three widely used datasets MNIST,CIFAR-10 and NUS-WIDE show that SDNMSH achieves better retrieval performance, compared with the state-of-the-arts. [ABSTRACT FROM AUTHOR]

To solve the problem of inaccurate description and unreliable relation inference in group activity recognition, this paper focused on constructing a scene relationship graph for three aspects: individual, group, and scene, and proposed a scene relationship graph network（SRGN） for group activity recognition. This method included a feature extraction module, a scene relation graph inference module, and a classification module. The feature extraction module extracted individual features, group features, and scene features by convolutional neural network. To fully explore the impact of scene on individual and group descriptions, the scene relation graph inference module learnt individual features and group features by building individual-scene and group-scene relationship graphs in a two-branch framework. Scene graph inference took into account the influence of individual on group and introduced a cross-branch module. It used the classification module to classify individual features and group features for prediction. The experimental results show that the group recognition accuracy of the proposed method on volleyball and collective activity data sets is improved by 1. 1% and 0. 5%,respectively. It verifies the validity of the scene graph in describing individual feature and group feature. [ABSTRACT FROM AUTHOR]

This paper proposed a natural language sentence matching method that combined high-level and low-level semantic information to solve the problems about current natural language sentence matching method fail to integrate common semantic information and it is difficult to capture deep-semantic information. First of all, the method used pre-trained word vector GloVe and character-level word vector to obtained the word embedding representation of sentence P and sentence Q. Secondly, this paper encode red P and Q with bidirectional LSTM, then it contained low-level semantic information through preliminary fusion of P and Q. Thirdly, this paper calculated bidirectional attention between P and Q, then spliced them together to get semantic representation, afterwards it calculated its self-attention to obtained high-level semantic information. Finally, this paper used a heuristic fusion function to fuse the low-level semantic information with the high-level semantic information to obtain the final semantic representation, and it used a convolutional neural network to prediction answers. This paper evaluated the proposed model on two tasks, such as recognition textual entailment, paraphrase recognition. This paper conducted experiments on the SNLI dataset and the Quora dataset. The results show that the accuracy of the proposed algorithm on the SNLI test set is 87. 1%, and the accuracy of the Quora test set is 86. 8%, which verifies the effectiveness of the algorithm in the task of natural language sentence matching. [ABSTRACT FROM AUTHOR]

Facial local detail information plays an important role in facial expression recognition（FER）.However, most of the existing methods only focus on the high-level semantic information of facial expressions, while ignoring the fine-grained information of local facial regions. To solve this problem, this paper proposed a deep convolutional neural network fusing local feature and two-stage attention weight learning（FLF-TAWL）,which could adaptively capture important facial regions to improve the effectiveness of facial expression recognition. The FLF-TAWL model was composed of a dual-branch framework, one branch extracted local features from image blocks, and the other branch extracted global features from the entire expression image. Firstly, this paper proposed a two-stage attention weight learning strategy. In the first stage, it roughly learned the importance weights of global and local features, in the second stage, it further refined the attention weight, and fused the local and global features. Secondly, the model used a region-biased loss function to encourage the most important regions to obtain higher attention weights. Finally, this paper carried out extensive experiments on FERPlus, Cohn-Kanada（CK+） and JAFFE datasets to obtain accuracy rates of 90.92%,98.90% and 97.39% respectively. The experimental results verify the effectiveness and feasibility of the FLF-TAWL model. [ABSTRACT FROM AUTHOR]

Copyright of Application Research of Computers / Jisuanji Yingyong Yanjiu is the property of Application Research of Computers Edition and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

Copyright of Application Research of Computers / Jisuanji Yingyong Yanjiu is the property of Application Research of Computers Edition and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

This paper proposed a self-encoding salient region extraction algorithm based on multi-layer convolutional to solve the problem that the existing convolutional neural network wasn’t sufficient in extracting image features, which led to the poor effect of salient extraction. When convolutional networks were used to extract image features, the shallow convolution features generally extracted the detailed features of the image, such as color, texture and location features, and the deep convolution features were generally referred to the semantic features of the image. On the coding layer, this algorithm downsampled the shallow convolution features and fused them into the deep convolution features, and upsampled the deep convolution features to the shallow convolution feature. Experiments show that this method can greatly improve the coding quality. In the decoding, it also fused the convolutional features by encoding to obtain the lost information in the decoding and also obtained a better decoded image. In addition, this paper designed a layer-by-layer supervision method to guide the training of the decoding layer, which used the standard region extraction map by down-sampling as the standard image for supervision training. The experimental results show that this algorithm can increase the F-measure by 0.071 on the basis of PAGRN, and reduce the MEA by 0.031 on average. [ABSTRACT FROM AUTHOR]

At present, there are few researches on aspect level sentiment analysis based on convolutional neural network and recurrent neural network, which take into account the syntactic structure of sentences and the syntax distance of words, and convolutional neural network and recurrent neural network can not effectively deal with the data of graph structure. To solve these problems, this paper proposed an aspect level sentiment classification model based on distance and graph convolution network. Firstly, this paper designed a two-layer bi-directional long short-term memory network with residual connection for the model to extract the context information of sentences. Then, the model obtained the weight of syntax distance of words and constructed the adjacency matrix from the syntactic dependency tree. Finally, the model used the graph convolution network to extract the sentiment features from context information, syntax distance weight and adjacency matrix. The experimental results show that the model is effective and can get better performance. [ABSTRACT FROM AUTHOR]

To solve the problem of combustion state recognition in solid oxide fuel cell (SOFC) combustor, this paper proposed a combustion state recognition method based on attention mechanism and image feature pyramid. The method adopted the adaptive gamma correction with weighting distribution (AGCWD) to standardize the input images. It combined two single full connections with 1×1 convolution to replace the squeeze and excitation structure, and proposed a hybrid attention structure combined with spatial attention structure to enhance the ability of feature extraction. To provide multi-scale information communication capability, it constructed the multi-scale bidirectional fusion pyramid by means of bidirectional computation and multi-scale fusion. The experimental results show that the proposed method reaches 99.22% accuracy under the premise of 3.98 M parameters and 397 M floating point operations (FLOPs), and effectively identifies the combustion state in SOFC combustor. [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]

The severe occlusion and scale variation in the cotton field reduce the accuracy of the object counting algorithm greatly. To solve above problems, this paper presented a cotton boll counting algorithm based on context multi-scale fusion. This algorithm consisted of the context module of the pyramid structure and the fusion convolutional neural network. Firstly, it adopted global context and local context modules to encode the context information in the cotton images, and then used multicolumn feature conversion module to map the input image into high-dimensional feature . Finally, it combined the context information with the high-dimensional feature through the fusion convolutional neural network, for achieving high-precision cotton boll counting and generating high-quality cotton boll density map. In addition, this paper conducted experiments on the cotton boll dataset from two angles of close-range and ground-air observation. The comparative experimental results show that adopting context information can effectively improve the accuracy of cotton boll counting, and the counting errors MAE and MSE have decreased by 27 . 3 and 29 . 4, respectively. [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]

In order to solve the problems of uneven population distribution and different target scales affecting the crowd counting in public places, this paper proposed a novel crowd counting model based on image field division. Firstly, this model divided the image scene into two parts : the near and far field of vision area. For the near field of vision area, it used the YOLO based network for p edestrian d etection and added scene constraints to avoid repeated counting in the near and far field of vision. For the far field of vision area, it used the improved MobileNets to extract the population density distribution characteristics, and introduced the super-resolution reconstruction module to improve the quality of the population density map . Finally, it obtained the number of the c rowd in the whole image by calculating the sum of the two. This paper tested the proposed model on Shanghai Tech and Mall d atasets. The results show that the model has a significant improvement in accuracy and robustness and prove the model is feasible. [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]

Common models of text classification are mostly constructed with recurrent neural network and convolutio-nal neural network in a stacked way. Although this stacked structure can extract more high-dimensional and deeper semantic information, a part of the effective feature information is also dropped when different structures are connected . In order to solve the above problem, this paper proposed a classification model based on dual-channel word vectors, and the model used a shallower structure with attention-mechanism-based Bi-LS TM and CNN to extract features of text representation effectively. In addition, this paper presented a new method to characterize text into two forms, forward and backward, and used CNN to extract feature information of the text . By conducting classification experiments on two different five -classification datasets and comparing with a variety of benchmark models, it verifies that the model is effective and the results show that this model is superior to the other models with stacked structure. [ABSTRACT FROM AUTHOR]

In recent years, single device sound scene classification has achieved good results, however, progress in multi-device sound scene classification has been slow. To solve the problem of large differences in the number of multi-device samples, this paper proposed a pairwise feature fusion algorithm. By calculating the differences in the spectrogram for each pair of paired samples and averaging these differences after accumulation, obtaining the average spectral characteristics of each device for the conversion of device samples. The algorithm effectively improved the generalization ability of the model while increasing the number of device samples. Meanwhile, in order to obtain global information, it proposed a lightweight attention module, which could make the model focus on the training of the whole sound sequence information on the basis of reduced computation by performing self-attention operations on the input features after compression in the frequency domain. The experimental results show that the proposed algorithm has better advantages in terms of model size and classification accuracy compared with other methods. [ABSTRACT FROM AUTHOR]

In order to solve the problems of high computational complexity, large computational cost and large number of parameters, this paper proposed the parallel group convolution algorithm based on the domestic SW26010P multi-core processor. The core idea was to use the unique data layout, through the multi-core mapping processing, parallel computing. Experimental results show that compared with single-core serial algorithm, the proposed parallel group convolution algorithm can achieve the highest speed-up ratio of 79.5 and the maximum effective computing power of 186.7MFLOPS. After data parallel optimization of the parallel group convolution algorithm by SIMD instruction, the algorithm obtains the highest speed-up ratio of 10.2 compared with the parallel group convolution algorithm before optimization. [ABSTRACT FROM AUTHOR]

Considering the complex characteristics of low contrast, blurred boundaries, and irregular shapes in the lesion area and surrounding mucus in colorectal polyp images, most of the existing algorithms cannot achieve high-precision segmentation of colorectal polyps. In view of the above difficulties, this paper proposed a colorectal polyp segmentation algorithm combining PVTv2 and multi-scale boundary polymerization. Firstly, using PVTv2 to extract the lesion features in the intestinal polyp image layer by layer, to solve the problem that the traditional convolutional neural network has insufficient ability to extract the features of the lesion area. It constructed a multi-scale context space awareness module at the network skip connection. Then, it designed a multi-scale extrusion adaptation fusion module to aggregate the feature information of different scales to reduce the semantic difference of each scale feature. Finally, in order to further strengthen the edge detail features recognition ability, creatively constructed residual axial double boundary refinement module. The algorithm verified by a large number of experiments on the Kvasir-SEG dataset and the CVC-ClinicDB dataset, the similarity coefficients were 93. 29% and 94. 52%, and the average intersectional merge ratio was 88. 36% and 89. 88% respectively. The experimental results show that the segmentation accuracy of the proposed model has been greatly improved for the complex lesion area and the blurred lesion boundary. [ABSTRACT FROM AUTHOR]

Current intent recommendation researches extract users' intents are flat, which ignore the hierarchical relationship between intents. In order to solve the above problem, this paper proposed the hierarchical intent disentangling for graph convolution neural network recommendation model ( HIDR), which divided the user-item interaction graph into multiple dynamic interaction sub-graphs to depict the hierarchy of user intent from fine-grained to coarse-grained. Firstly, in each intent interaction sub-graph, the model adaptive fused information from high-order neighborhoods according to the high-order connectivity of nodes, and extracted the fine-grained intent representations of users by disentangling. Then, it built coarse-grained intents based on the similarities between the fine-grained intents in low-level relationships and high-level network super-nodes to explicitly model the intent hierarchy from fine-grained to coarse-grained. Finally, it aggregated the high-quality representations of user and item for inner product predictive and iterative optimization by the hierarchical intent vectors. The experimental results on the two datasets of Gowalla and Amazon-book show that the recall rate of HIDR is improved by an average of 10. 82% and 6. 63%, and the normalized damage cumulative gain increases by an average of 14. 65% and 9. 63%, and the precision increases by an average of 10. 46% and 7. 73 %, respectively, compared with optimal benchmark model CLSR. [ABSTRACT FROM AUTHOR]

Most of the existing steganalysis methods have weak generalization ability and cannot effectively detect unknown steganalysis algorithms, which makes the accuracy of their classification greatly reduced in the practical application process. To solve this problem, this paper proposed an image steganalysis method based on group convolution and snapshot ensembling（Snapshot Ensembling steganalysis network, SENet）. Firstly, residual convolution block and group convolution block are used to extract the features of the image. Secondly, the model with the best performance is obtained as the snapshot model in each training period. Finally, the selected snapshot models are integrated to classify the images. This method uses the techniques of group convolution and snapshot ensembling to avoid the high training cost of traditional integration methods and the limited generalization ability of a single classifier. Experimental results show that this method can improve the accuracy of steganographic analysis model, and can effectively classify when the training set and test set are mismatched, and has high model generalization ability. [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 Transformer model cannot apply to mobile devices with limited computational resources due to its huge number of parameters and computational complexity, this paper proposed a mobile-friendly lightweight image recognition network called FuseNet. FuseNet utilized convolutional neural network to extract local feature information and self-attention to excel in modeling global information, and it integrated the features of both local and global representations into a single feature extraction module, which efficiently combined the advantages of the two different structures to achieve a high accuracy with a small model size. Experiments demonstrate that FuseNet with different model sizes can achieve good performance without using pre-training and it can well apply to mobile devices. FuseNet-B achieves 80. 5% accuracy with 14. 8 M parameters on ImageNet-1 K dataset, the performance exceeds the same volume of Transformer models and convolutional neural networks. [ABSTRACT FROM AUTHOR]

In the field of CXR image classification, current studies mostly use convolutional neural network and transfer learning technology. But in the process of rapidly constructing a network without considering the specificity of the CXR image. To solve the above problem, this paper proposed a novel XDense-RC-net. Based on the DenseNet model, this method improved the capability of feature extraction and feature fusion by introducing a new spatial attention mechanism in the densely connected layer, Optimized Transition module of DenseNet by using two different pooling strategies to enhance the noise immunity of the model. XDense-RC-net used the chest X-ray14 (multilabel and 14-class) and COVIDx (single label and 3-class) datasets for validating. In the multi-label classification experiments, the average AUC score reached 0.854, which is 0.109 higher than the benchmark method. In the single-label classification experiments, the average accuracy reached 96.75%, which is 7.75% higher than the baseline model. The results show that XDense-RC-net improves the accuracy of CXR image classification and can generalize to multilabel and single-label classification tasks. [ABSTRACT FROM AUTHOR]

In order to improve the accuracy of text classification and solve the problem of insufficient utilization of node features by text graph convolution neural network, this paper proposes a new text classification model, which integrates the advantages of text graph convolution and Stacking integrated learning method. The model first learns the global expression of documents and words and the grammatical structure information of documents through text graph convolution neural network, and then secondary learns the features extracted by text graph convolution through integrated learning, so as to make up for the insufficient utilization of text graph convolution node features, and improve the accuracy of single label text classification and the generalization ability of the whole model. In order to reduce the time consumption of ensemble learning, the fusion algorithm removes the k-fold cross verification mechanism in ensemble learning. The fusion algorithm realizes the correlation between text graph convolution and stacking integrated learning method. The classification effect on R8, R52, Mr, Ohsumed, 20 ng and other data sets is improved by more than 1.5%, 2.5%, 11%, 12% and 7% respectively compared with the traditional classification model. This method performs well in the comparison of classification algorithms in the same field. [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]

At present, aspect-based sentiment analysis based on graph convolution neural network (GCN) integrates the grammatical structure of sentences. It can effectively solve the problem of inaccurate long-range lexical dependence. However, GCN has unnecessary complexity and redundant computing. In addition, it ignores the relative position relationship between attributes and context. This paper proposed a new model to solve these problems. Firstly, the model established a bidirectional GRU layer. Secondly, it used position-aware transformation to increase the importance of context words close to aspect words, and then reduced the complexity by removing nonlinearity and folding the weight matrix between successive layers. Thirdly, it fused the specific mask layer to realize the single-layer MASGC (masked simple graph convolutional networks) structure, and then generated a new attention mechanism based on retrieval context. Finally, it gave the classification results through the full connection layer. The model carries out a large number of experiments on five data sets. The experimental results show that it has higher accuracy and less training time. [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]

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]

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]

In order to solve the problems of insufficient feature extraction of sEMG end-to-end gesture recognition and low accuracy of multi-gesture recognition, this paper proposed a multi-stream convolutional EMG gesture recognition network model fusing attention mechanism. The proposed model generated sub-myoelectrograms from multi-channel time-domain sEMG through sliding windows and used a multi-stream convolutional neural network to fully extract the semantic features of each sampled channel sEMG, and then aggregated them to obtain rich multi-channel gesture semantic features. At the same time, the proposed model calculated the attention distribution map of semantic features from temporal and feature channel dimensions to strengthen the useful features and weaken the useless features, further improved the accuracy of multi-gesture recognition. The experiment used the Ninapro dataset for training and testing and compared with the mainstream electromyogram recognition model. The experimental results show that the proposed model has better performance in recognition accuracy, which proves the effectiveness of the proposed model. [ABSTRACT FROM AUTHOR]

The size of the receptive field of the convolutional neural network is related to the size of the convolution kernel. And the traditional convolution uses a fixed-size convolution kernel, which limits the feature perception ability of the network model. In addition, due to the parameter sharing mechanism of the convolutional neural network, it used the same feature extraction method for all pixels in the spatial region. However, there are differences in the distribution of noise signals and clean speech signals in the noisy spectrogram, especially in the complex noise environment, the general convolution method is difficult to achieve high-quality speech signal feature extraction and choosing. In order to solve the above problems, this paper proposed a multi-scale region adaptive convolution module, which used multi-scale information to improve the feature perception ability of the model and automatically allocated the area adaptive convolution achieve and improved the denoising ability of the model. The experiments on the TIMIT public datasets show that the proposed algorithm has achieved satisfactory results in the metrics of speech quality and intelligibility. [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]

Aiming at the problems that the current edge detection algorithms rely on the FC layer too much, making the edge line rough and the improper setting of the loss function, which make the gradient disappear and the main characteristics loss. To solve the two problems, this paper proposed a cross-layer fusion edge detection algorithm based on convolutional neural network . It used 1 x 1 multiple convolution kernels gradient method to reduce the dimension, and completed the acquisition of low level and high-level feature objects of horizontal and vertical images. Then, the convolutional flowing direction was from top to bottom and from left to right, so that the loss function of each layer could be propagated forward and backward smoothly. Finally, it refined the image edge feature by cross-layer fusion. Experimental results show that ODS F-measure is 0. 806 on BSDS500 dataset, which is close to human average visual perception. [ABSTRACT FROM AUTHOR]

In order to solve the problem that the existing sequence recommendation algorithms only use short-term sequential behaviors to make recommendations, but fail to fully consider the long-term preference of users and the deeper relationship be tween items, this paper proposed a sequential recommendation algorithm CSALSR that it integrated self-attention mechanism and long-term and short-term preference. This model embedded the user and the item into the vector respectively, and regar ded the pairwise encoding of the item in short-term interaction sequence as a three-way tensor. Then, it added self-attention mechanism on the three-way tensor and used convolutional neural network to extract the deeper connection between items from the order pattern of users. At the same time, considering the long-term preference of users, it took the embedding of similar users as a supplement to enhance the user representation. Experimental results show that the proposed method on MovieLens IM and Gowalla data sets is superior to other methods in precision@N.recull@ N and MAP. [ABSTRACT FROM AUTHOR]

Based on convolution neural network, this paper proposed a new asymmetric two branch underwater biological classification model to solve the problems of low visibility, poor illumination conditions and no obvious differences among species in the underwater environment. In the model, the interactive branch used different convolution neural network to extracted local features and interacted with local features through the interactive module to enhance the classification model. Convolutional neural network branch could effectively learn the global characteristics of the target and made up for the global information ignored in the interactive branch. Finally, this model obtained 98 . 9%, 98 . 3% and 97 . 9% of the accuracy on the three data sets of Fish4-Knowledge (F4K), Eilat and RAMAS, which were significantly improved compared with the previous methods. Visual interpretation also verifies that the model could effectively capture local features and eliminated the background influence. Finally, it shows that the model has good classification performance in underwater environment. [ABSTRACT FROM AUTHOR]

In order to solve the problems of network model complexity and low identification rate, this paper proposed a person re-identification( Re!D) method based on global feature stitching. Firstly, it extracted the global features using convolutional neural network (CNN). Secondly, it stitched the features from different convolution layers together to complement the feature information. Finally, it convoluted again to obtain the features with high representation ability. In network training stage, it combined the cluster loss with label smoothing loss, and adopted random erasing augmentation (REA) as well as pooling step reduction techniques. It conducted extensive experiments to evaluate the performance of this proposed method on Marketl501, DukeMTMC-re!D, CUHK03 and MSMTl 7 benchmark datasets. Results show that the proposed method outperforms other state-of-the-art techniques, for instance, the Rank-1 and mAP on Marketl501 are 95. 9% and 94. 6%, respectively. [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]

Most of the relation extraction approaches could not learn the long distance dependence information from the long sentences with entity co-occurrence. This paper proposed a new relation extraction model to solve this problem. It was based on the recurrent convolutional neural network and the sentence-level attention mechanism. It used the Bi-GRU neural network to learn context vectors for words. And it adopted the piecewise maximum pooling method, which could obtain fine grained features. Experimental results on the NYT dataset demonstrate that this proposed method outperforms the baseline systems. [ABSTRACT FROM AUTHOR]