Search

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.)

Implementing fast and accurate hand detection on embedded devices mainly face two challenges. Firstly, it is difficult for complex deep learning networks to achieve real-time hand detection. Secondly, the complexity of the scene leads to a decrease in the accuracy of hand detection algorithms based on RGB color images. Unlike mainstream RGB image based detection technologies, this paper adopted a lightweight hand detection algorithm based on ToF infrared images to attain precise and swift hand detection within the infrared images. Firstly, this paper gathered 22 419 static infrared images using this self-engineered equipment, thereby establishing an infrared dataset tailored for hand detection. Subsequently, it enhanced a general object detection algorithm to create a lightweight hand detection network known as Retina Hand, using two different lightweight networks, MobileNetV1 and Shuffle NetV2, as the backbone network of the model. Furthermore, this paper proposed an attention-enhanced feature pyramid structure called Attention-FPN. This structure integrated attention mechanisms to enhance the detection process. Ultimately, this paper conducted comparative experiments on the infrared dataset against conventional methods to validate the effectiveness of the method. [ABSTRACT FROM AUTHOR]

HD maps can provide accurate road information and rich semantic information, so that the automatic driving system can guide the vehicle to drive correctly. HD maps usually rely on manual labeling. The existing automatic labeling methods have low recognition accuracy in autonomous driving scenarios, resulting in low efficiency of HD map labeling. In order to solve this problem, this paper proposed a new semantic segmentation method for automatic labeling of HD maps called MapFormer. This paper proposed a multi-level feature fusion module, which enabled the model to aggregate different levels of detail and semantic information. Additionally, it proposed a novel boundary-decoupled joint decoder to improve the model's ability to handle intercategory boundaries. Experiments on the bird's-eye view dataset show that the proposed model not only achieves excellent performance in segmentation accuracy, but also has a clearer handling of category boundaries. The mIoU of MapFormer is 55.82%, which is higher than the SegFormer of 1.03%. The proposed method can improve the HD map labeling efficiency and labeling automation rate. [ABSTRACT FROM AUTHOR]

To address the problem of insufficient recommendation performance in session recommendation, this paper proposed a disentangled graph neural network model (GDST-GNN) using global graph diffusion and spatio-temporal awareness. Specifically, the model firstly constructed a global collaborative graph based on all sessions from a global perspective, and then employed graph diffusion as the message propagation paradigm for global representation learning of items to capture global information be- yond the current session. For the representation learning of the current session, this paper designed a disentangled spatio-temporal gated network modeling the complex transition pattern and temporal dependency pattern of items in the session, and then fused the learned global and local representations factor by factor. In addition, this paper employed a self-supervised task to achieve performance enhancement of the model. Finally, it generated session representations through attention networks to achieve accurate recommendation of items. Extensive experiments on four real-world datasets validate the effectiveness of the proposed model. [ABSTRACT FROM AUTHOR]

For the problem of image degradation caused by raindrops attached to the lens or glass surface, this paper proposed a multi-stage progressive image raindrop removal method which divides the raindrop removal process into multiple stages to easily implement. First, the multi-scale fused encoder-decoder network at each stage designed to learn raindrop features. The constructed multi-scale dilated convolutions with gated recurrent units refined the internally transferred spatial features. In addition, the no dimensionality reduction channel attention module extracted channel information under specific spatial features. Finally, this paper adopted cross-stage feature fusion mechanism to strengthen the information exchange between the parts of each stage. Adding horizontal connections between the encoder-decoder networks in each stage to realize the horizontal transfer of feature information. And adding a supervised attention module to enhance the information transmission between different stages. The raindrop removal is progressively achieved. Experiments show that the proposed method can effectively remove raindrops. [ABSTRACT FROM AUTHOR]

Aiming at the problem that feature extraction is inaccurate and the reconstruction image is not natural enough in blind super-resolution reconstruction, this paper proposed an image blind super-resolution reconstruction based on degradation aware and sequence residuals. This paper proposed a mini-residual group combined degeneration aware and sequence residuals as the backbone network. Then the method constructed a symmetrical enhanced multi-scale residual block. In the image reconstruction part, this paper used the bottleneck attention module and the sub-pixel convolutional module to emphasize the multi-dimensional elements of the image. Finally, the method made a global residual connection. Compared with the current representative algorithm DASR, experiments show that the PSNR and SSIM of the proposed algorithm are improved 0.145dB and 0.0014 on Set14×2, and the PSNR of the proposed algorithm is improved 1.898dB and 0.252dB on Set14×3/4, respectively. The proposed algorithm achieves better performance than several current image super-resolution algorithms on five standard test sets. [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.)

Aiming at the problem of insufficient feature extraction ability of deep clustering algorithm on multivariate time series data( MTS), This paper proposes a new deep clustering structure model (MDTC. In order to extract the key features of MTS and realize dimensionality reduction, one-dimensional convolution learns the attribute and temporal dimension feature representation of MTS and the autoencoder structure composed of network layers such as recurrent neural network; To improve the model’s ability to represent temporal features, this paper proposes MCBAM temporal attention module, which is used to enhance the representation features of different time periods in the MTS sequence. This paper conducts experiments on nine publicly available UEA multivariate time series datasets, compared with other autoencoders, the autoencoder structure of the model improves by 2%-9% on 7 data sets; Compared with other attention modules, the MCBAM module of the model improves by 0.3%-2% on the six datasets. Experiments show the effectiveness of the MDTC model structure and MCBAM module, and the model has excellent performance compared with other clustering algorithms. [ABSTRACT FROM AUTHOR]

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]

Video distortion mainly comes from the degradation of video quality caused by spatial and temporal distortion. Aiming at these two kinds of video quality degradation, this paper proposed a full-reference video quality assessment method STPFVQA combining spatio-temporal features and visual perception. Firstly, the method extracted spatial perceptual features from reference video and distorted video using ResNet50 convolutional network. Secondly, the method feeded the extracted spatial perceptual features into transformer’s encoder and decoder to construct the serialization relationship of the video, and at the same time to explore the effect of distortion on the video serialization relationship by comparing the reference video and distorted video Then, the method sent the output of the transformer to the prediction header to form a frame-level score Finally, in order to simulate the hysteresis of human visual system perception, this paper obtained the video quality score from the short-term, long-term and global memory effects. To verify the feasibility of the method, this paper conducted experiments on four datasets, respectively LIVE,IVC-IC,CSIQ,and IVPL. The experimental results show that the proposed method is more consistent with the perception of the human visual system. Compared with the state-of-the-art serial dependence model（SDM） on the IVC-IC and CSIQ datasets, the SROCC is 2.6% and 3.1% higher, the KROCC is 6.1% and 7.9% higher, and the PLCC is 2.3% and 5.5% higher. [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]

Long-tail characteristic of data has always been a great challenge in solving computer vision problems. To solve the difficulties brought to instance segmentation, image re-sampling, a simple and efficient method has always been used. However, there may be many classes in one image, it’s hard to balance different classes on the data level. This paper proposed a object-centric post-processable memory bank method. Firstly, it introduced a memory bank model and set up an object-based storage policy to address the number imbalance of categories. Then, in order to increase the predicted score for tail classes and common classes, it used a post-processing calibration to adjust the confidence score of each class. This paper verified the effectiveness of the proposed method by experiments on LVIS dataset. The accuracy can be improved by 2.2% compared to EQL method. [ABSTRACT FROM AUTHOR]

COVID-19 is spreading rapidly around the world. In order to diagnose it quickly and accurately and thus block the chain of epidemic transmission, the study proposed a deep learning-based classification network DLDA-A-DenseNet. Firstly, DenseNet-201 combined deep layer dense aggregation to aggregate feature information at different stages to enhance its ability to identify and localize lesions. Secondly, this paper proposed efficient multi-scale long-range attention to refine the aggregated features. Moreover, this paper used a balanced sampling training strategy to eliminate the bias for the class imbalance problem of CT image dataset. Testing on the China consortium of chest CT image investigation dataset, the method improved 2.24%,3.09%,2.09%,2.60% and 3.48% in accuracy, recall, precision, F1 score and Kappa coefficient compared with DenseNet-201,and achieved an optimal accuracy of 99.50% on COVID-CISet image dataset. The results show that the proposed COVID-19 CT image classification method can fully extract the lesion features of CT slices compared with other methods, and has higher classification accuracy and good generalizability. [ABSTRACT FROM AUTHOR]

Collaborative intrusion detection system (CIDS) can detect distributed cooperative attacks and deal with large-scale network intrusion. It has the advantages that traditional intrusion detection systems do not have. How to improve the detection performance while achieving decentralization is an important research topic. Through reviewing the research achievements of CIDS in recent years, this paper expounded the research progress and problems in detection methods, data aggregation, privacy protection and trust management, analyzed the challenges faced in decentralization, discussed the future development direction of CIDS based on blockchain technology. Finally this paper looked forward to the application prospects of CIDS in cloud computing, the Internet of Things and other emerging fields. [ABSTRACT FROM AUTHOR]

In recent years, various human action recognition algorithms have achieved excellent recognition performance based on a large number of labeled samples. However, in practical applications, acquiring training samples and their corresponding labels is time-consuming and laborious, which limits the actual implementation of the algorithm. This paper summarized the deep learning algorithms for action recognition under weak supervision and few-shot learning. Firstly, in the case of weak supervision, it classified and summarized the semi-supervised action recognition methods and unsupervised domain adaptation video action recognition methods separately. Then, it reviewed video action recognition algorithms based on few-shot learning in detail. Further, it summarized the relevant human behavior datasets and analyzed and compared the performance of various relevant video action recognition algorithms on these datasets. Finally, the paper summarized the full text and discusses the future development trend of human action recognition. [ABSTRACT FROM AUTHOR]

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

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.)

The radar echo extrapolation method is widely used in rainfall forecasting. Addressing the issue of insufficient prediction accuracy in radar echoes, this paper proposed a deep learning model DIPredRNN based on recurrent neural networks. This model combined long-term temporal and channel information by introducing a dual attention mechanism of space and channel, improved the long-term dependence of time memory. By introducing an interactive framework of hidden states and inputs, it retained more features and improved the short-term dependence of temporal memory. This model was experimentally compared with classical models and many advanced models on the HKO-7 and Sichuan datasets. The model achieved the best results in comparing multiple indicators such as extrapolated images, MSE, SSIM, CSI-30~50 dbz. The experiment proves that the proposed DIPredRNN network improves the radar echo prediction performance and has advanced performance. [ABSTRACT FROM AUTHOR]

The majority of current machine learning and deep learning solutions for ECG-based emotion recognition utilize fully-supervised learning methods. Several limitations of this approach are that large human-annotated datasets and computing resources are required. Furthermore, the feature representations learned by fully supervised methods tend to be task-specific with limited generalization capability. In response to these issues, this paper proposed an approach based on contrastive lear-ning for ECG-based emotion recognition, which consisted of two steps, such as pre-training and fine-tuning. The goal of pre-training was to learn representations from unlabeled EGG data through contrastive learning. Specifically, it designed two simple and efficient ECG signal augmentation methods, and used these two views to learn robust temporal representations in the time contrastive module, followed by learning discriminative feature representations in the context contrastive module. Fine-tuning used labelled data to learn emotion recognition. Experiments show that the proposed method has reached the maximum accuracy on three public ECG-based emotion recognition datasets. Additionally, the proposed method shows high efficiency under the semi-supervised settings. [ABSTRACT FROM AUTHOR]

As a popular research direction in the field of artificial intelligence, Facial Expression Recognition (FER) is the most direct and effective way for computers to access human emotions. It occupies a key technical position in human-computer interaction, intelligent medical care, fatigue driving, and other R＆D topics. In order to meet the application requirements of high recognition rate, the structure of FER deep learning network becomes more and more complex, occupying a large amount of computing resources and storage space, which seriously affects the real-time requirements of the algorithm. This paper focuses on the problem of how to guarantee the accuracy of the model while effectively improving its computational speed. Firstly, it introduces the important datasets for expression recognition using lightweight networks; Secondly, it analyzes the classical lightweight network models used for facial expression recognition; Thirdly, it describes the principles, characteristics, and applicable scenarios of the main network lightweighting methods; Finally, it summarizes the problems and challenges of lightweight networks in facial expression recognition research and looks forward to the future research direction. [ABSTRACT FROM AUTHOR]

The layout of document images is complex and distribution of object sizes is uneven, currently, most of detection methods ignore multimodal information and global dependencies. Therefore, this paper proposed a multi modal document object detection method based on vision and text. Firstly, this method explored the fusion strategy of multimodal features. In order to utilize textual features, it converted text sequence information of the image into two-dimensional representation. After the initial fusion of text features and visual features, it input the fused features to backbone network to extract multiscale features, and repeatedly integrated textual features during the extraction process, so as to realize deep fusion of multimodal features. Next, to ensure the detection accuracy of small and large objects, this paper designed a pyramid network. The lateral connection could concatenate feature maps of the same spatial size from the bottom-up pathway and the top-down pathway in channel, so as to achieve the propagation between high-level semantic information and low-level feature information. The experimental results on large public dataset PubLayNet show that the detection accuracy of this method reaches 95. 86%,and it has a higher accuracy than other methods. This method not only realizes the deep fusion of multi modal features, but also enriches the fused multimodal feature information, and it has good detection performance. [ABSTRACT FROM AUTHOR]

As a new research hotspot in deep learning, intermediate deep feature compression has gotten a great deal of attention and has been applied to edge-cloud intelligent collaboration. This paper summarized the current state of research on the intermediate deep feature compression and analyzed the problems in the current methods. Firstly, it introduced three kinds of intermediate deep feature compression from aspects of image/video coding framework, channel bit allocation and network units. Then it did the comparisons of the data set performance between the three intermediate deep feature compression. Finally, this paper discussed the existing challenges and solutions in intermediate deep feature compression and looked forward to the feature trends of intermediate deep feature compression. [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]

Object surface defect detection technology is an important subject in the field of industrial quality inspection, which is of great significance to industrial production. This paper mainly summarized the surface defect detection techniques based on machine vision in recent years. Firstly, it listed several application scenarios of defect detection in the industry. Secondly, from the perspective of feature extraction and classification algorithms, it expounded on traditional machine vision methods. Then, this paper focused on the classical neural network structure and the latest development of defect detection algorithms and introduced two commonly used optimization methods of the defect detection algorithm. Finally, it analyzed three major challenges in the field of defect detection; real-time problems, small sample problems and small target problems. The purpose was to provide useful references and context for the research of industrial surface defect detection. [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.)

The current research on the application of artificial intelligence techniques to the field of system architecture is promising, especially the research on applying deep learning to data prefetching in multicore architectures has become a research hotspot at home and abroad. This work studied the cache prefetching task based on deep learning and defined the deep learning cache prefetch model formally. Based on the introduction of current common multi-core cache architectures and prefetching techniques, this paper comprehensively analyzed the design ideas of existing typical cache prefetchers based on deep learning. The application of deep learning neural network in the field of multicore cache prefetching mainly adopts machine learning methods such as deep neural network, recurrent neural network, long and short-term memory network and attention mechanism. A comprehensive comparative analysis of existing deep learning-based data prefetching hierarchical neural models reveals that deep learning-based multicore cache prefetching techniques still have certain computational cost, model optimization, and practicality. In the future, there is still much room for research exploration and development prospect in adaptive prefetching models and the practicality of neural network prefetching models. [ABSTRACT FROM AUTHOR]

Atmospheric turbulence is a crucial factor that affects the quality of long-distance imaging. Though current deep learning models can effectively suppress geometric displacement and spatial blurring caused by atmospheric turbulence, such models require a large number of parameters and computational resources. To tackle this problem, this paper proposed a lightweight atmospheric turbulence degraded image restoration model based on generative inversion that entailed three core mo-dules: the DeBlur module, the remove shift module, and the turbulence regeneration module. The DeBlur module used high-dimensional feature mapping blocks, detail feature extraction blocks, and feature compensation blocks to suppress image blurring caused by turbulence. The remove shift module compensated for pixel displacement caused by turbulence using two convolutional layers. The turbulence regeneration module regenerated turbulence degraded images through convolutional operations. In the DeBlur module, it designed an attention-based feature compensation module that integrated the channel attention mechanism and the spatial mixed attention mechanism to focus on essential detail information in the image during training. The proposed model achieved peak signal-to-noise ratios of 19.94 dB and 23.51 dB, and structural similarity values of 0.688 2 and 0.752 1 on publicly available dataset Heat Chamber and self-built dataset Helen, respectively. Furthermore, it reduced the number of parameters and computational resources, compared to the current state-of-the-art(SOTA) method. The experimental results demonstrate the effectiveness of this method in restoring atmospheric turbulence degraded images. [ABSTRACT FROM AUTHOR]

The function of protein is crucial for understanding the mechanisms of cellular and biological activities, as well as for studying the mechanisms of diseases. Traditional experimental and sequence alignment methods are insufficient to support large-scale protein functional annotation when in the face of the rapid growth of sequence databases. For this situation, this paper proposed EGnet（Evolutionary Graph Network） model, which utilized the protein pre-training language model ESM2 and One-hot encoding to obtain the protein sequence encoding. The model integrated the coevolutionary information between residues, including PI（Paired Interaction） and SPI（Strong Paired Interaction）, through sequence self-attention and physical calculations. Subsequently, the two types of coevolutionary information and the sequence encoding used in inputs for a multi-layered cascaded graph convolutional network to learn the node features of the sequence encoding and achieve end-to-end protein function prediction. Compared with earlier methods, EGnet achieved better performance on the EC（Enzyme Commission） category labels in the ENZYME database, which reached 0.89 in the F-score and 0.91 in the AUPR. The results indicate that EGnet can achieve good performance by using only a single sequence to predict protein function, providing a rapid and effective method for protein function annotation. [ABSTRACT FROM AUTHOR]

Based on deep models, video action recognition typically involves sampling the input video and then extracting features from the obtained video frames to classify actions. Therefore, the video frame sampling method directly affects the effectiveness of action recognition. Aiming to sample key and effective features while enhanced the motion information in videos, this paper proposed a LGMeNet based on a feature-level sampling strategy. Firstly, it used a feature-level sampling module to uniformly select frames with the same motion information from the input data. Secondly, it employed a local motion feature extraction module to compute short-term motion features using a similarity function. Finally, it utilized a LSTM network in the global motion feature extraction module to calculate multi-scale long-term motion features. Experimental evaluations show that LGMeNet achieves accuracies of 97.7% and 56.9% on the UCF101 and Something-SomethingV1 datasets, respectively. The results of this study demonstrate the effectiveness of LGMeNet in enhancing action recognition and highlight its significance for further advancements in related research areas. [ABSTRACT FROM AUTHOR]

Aiming at the issue of complex 3D shape analysis and recognition, this paper presented a novel 3D graph convolu- tion classification method. It established a joint graph convolution learning mechanism of local geometry and global structure to provide both geometrical features and global context features, which effectively improved the robustness and stability of 3D data learning. Firstly, it constructed the local graph in spatial domain by farthest point sampling and K-nearest neighbor method, and designed a dynamic spectral graph convolution operator to extract local geometric features effectively. Meanwhile, it con- structed the global feature graph based on random sampling in the feature domain, and obtained the global structure context by spectral graph convolution. Furthermore, it established a weighted graph convolution network with an attention mechanism to achieve adaptive feature fusion. Finally, under the optimization of objective function, it improved the performance of feature learning effectively. Experimental results show that the proposed joint network learning mechanism, which combined local geo- metric features with global structure features, enhances the representation ability and discrimination of deep features, and ob- tains better recognition and classification performance compared with advanced methods. This method can be used for large- scale point clouds recognition, 3D shape reconstruction and data compression. It has important research significance and broad application prospects in robot, product digital analysis, intelligent navigation, virtual reality and other fields. [ABSTRACT FROM AUTHOR]

Aspect and opinion pair extraction aims to extract aspect and opinion items and match relations from a given sentence. However, related studies typically extract aspects and opinions independently without identifying the relationships. To identify the relationships of aspect and opinion item, this paper proposed a knowledge-augmented multi-task learning model for aspect and opinion pair extraction. First, it used the pre-trained language model to generate word vectors with semantic information for the text. In order to achieve the effect of knowledge enhancement, it used the masked attention mechanism to integrate the semantic information of the knowledge graph into the word vectors, and used the sequence labeling method based on the distance-based attention and conditional random fields to extract aspects and opinions. Finally, it matched the extracted aspects and opinions to predict the corresponding relationship. In order to strengthen the connection between the aspect and opinion extraction module and the matching module, the model adopted a shared coding layer to achieve joint training. In addition, in the training process, the matching module used the real labels as input, and used the result of the extraction module as input in the testing process. Finally, to demonstrate the effectiveness of the model, this paper used three general domain datasets for comparative experiments. The model achieves F, values of 66. 99%, 75. 17% and 67. 30% in aspect and opinion matching tasks respectively, and outperforms other comparative models. [ABSTRACT FROM AUTHOR]

To address the problem of poor recommendation accuracy in the session recommendation model, this paper proposed a capsule-based graph convolutional disentanglement for session-aware recommendation (CGCD) method.Specifically, it used disentangled learning technique to transition item embedding into factor embedding based on multiple sub-channels. Then, it used a capsule dynamic fusion strategy to aggregate different factors to obtain a new item embedding.In addition, it used a multi-head attention mechanism to assign weights to each item in the session. Finally, it aggregated the item embedding with other items in the current session according to the assigned weights, and then generated an accurate session representation to achieve item recommendation. Experiments on two publicly available real datasets show that the proposed model can improve the Pre@10, Pre@20, MRR@10 and MRR@20 by 5.17%, 2.99%, 6.56% and 2.94% on average of the recommended performance, which verifies the effectiveness and efficiency of the method in this paper. [ABSTRACT FROM AUTHOR]

In order to mitigate mode collapse when generative adversarial networks (GANs) synthesis images, this paper proposed maximizing middle modes and minor modes in generative adversarial networks (MMMGANs). First of all, MMMGANs defined the mode of the generated images with the identical label as the middle mode and the minor mode. The middle mode and the minor mode separately represented the collection of similar modes and the possible mode changes after learning the middle modes. Secondly, based on the above definition, this paper proposed maximizing middle modes and minor modes loss. Finally, on the premise that the distribution of generated images was possibly close to the real image, MMMGANs improved more than 90% of evaluation metrics and then enhanced diversity of generated images. Extensive experimental results show that the proposed maximizing middle modes and minor modes loss effectively alleviates mode collapse in two or more different types of tasks. [ABSTRACT FROM AUTHOR]

Bone age assessment is a common method to detect endocrine and growth abnormalities in children. But in deep learning methods, low-quality hand X-ray images reduce the final evaluation accuracy. To solve this problem, this paper proposed an alignment network that increases the area of interest in hand X-ray images. This network uses the Swin Transformer structure as the backbone network to learn image hand similarity and obtain affine coefficients and does not require large-scale hand annotation during the training process. In the bone age assessment network, for the improvement of efficient channel attention and spatial attention mechanism. This paper proposed dual-pool efficient channel attention and asymmetric convolution spatial attention method and combines these two methods in the form of dual attention and Xception network proposes DA-Xception. This paper tested the RSNA dataset, which achieved a mean absolute error of 5.37 months for this bone age assessment method. Compared with other deep learning methods, this method can fully extract features and optimize the evaluation results. [ABSTRACT FROM AUTHOR]

In order to solve the disadvantage that the ViT cannot change the input patch size and the input patches are all single-scale information, this paper proposed an image classification network based on Transformer called MultiFormer. MultiFormer embeds small patches with different scales of input at each stage into large patches with rich semantic information through the AWS（Attention With Scale） module; and captures local and global attention alternately through the GLA-P（Global-Local Attention With Patch） module, preserving both fine-grained and coarse-grained features during embedding. This paper designed MultiFormer-Tiny,-Small and-Base networks of three different sizes to achieve 81.1%, 82.2% and 83.2% Top-1 accuracy respectively in ImageNet image classification experiments, the latter two models improve by 3.1% and 3.4% compared to the same volume of convolutional neural networks ResNet-50 and ResNet-101; MultiFormer-Base offers a 2.1% improvement with far fewer parameters and computational effort than the ViT-Base/16 model, and without the need for extensive data pre-training. [ABSTRACT FROM AUTHOR]

This paper proposes MDC image enhancement method by using joint learning of side-and central-decoding features, which considers the problems of side decoding image enhancement and central decoding image enhancement at the same time, so it can realize better network training by optimizing central decoding and side decoding features through joint learning. First, considering side independent decoding and central joint decoding features for MDC, this paper proposed a network-sharing side low-resolution feature extraction network to effectively extract features from two-side decoded images with the same content and different details, while it design a residual recursive compensation network structure and apply it into both side and central low-resolution feature extraction network. Secondly, a multiple description up-sampling reconstruction network is designed, which adopts parameter sharing strategy for partial layers of network, which can reduce parameter number of network model and improve network generalization ability. Finally, multiple description central upsampling reconstruction network is proposed to perform deep feature fusion with two low-resolution side features and central features to enhance multiple description compressed images. A large number of experimental results have shown that the proposed method is superior to several image enhancement methods such as ARCNN, FastARCNN, DnCNN, WSR and DWCNN in terms of model complexity, objective quality and visual quality assessment. [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.)

The weakly supervised methods for image semantic segmentation using the image-level labels are a relatively popular research direction. The class activation maps generation approach is the most commonly used approach in these researches. Due to the sparsity of class activation maps, the accuracy of discriminative regions is generally low. To address these problems, this paper proposed an improved weakly supervised image learning method based on the Transformer network. Firstly, it introduced a spatial attention interaction layer to extend the coverage of the class activation maps. Secondly, it designed an attention adaptive module to guide the model to enhance the class response in weak regions. In particular, it constructed an adpative cross domain to improve the model classification performance during class generation. The method achieves the accuracies of 73.5% and 73.0% on the Pascal VOC 2012 validation sets and test sets, respectively. The experimental results prove that the refined Transformer network learning method can improve the image semantic segmentation performance of weakly supervised method. [ABSTRACT FROM AUTHOR]

For the beamforming design of intelligent reflector assisted large-scale MISO systems, most of the current researches are basing on the fully known channel state information, and do not consider the coupling relationship between the active beamforming matrix at the base station and the reflection coefficient matrix of intelligent reflector. The intelligent reflector assisted channel is a cascade channel, which is difficult to estimate and has high pilot frequency overhead. In view of this, this paper proposed a hybrid intelligent reflective surface architecture containing active and passive components. Firstly, based on the sparse characteristics of millimeter-wave channels, it used the compressed sensing algorithm to estimate the channels, and designed a two-stage convolutional network framework integrating attention mechanism to jointly optimize the reflection coefficient matrix and beamforming matrix with the aim of maximizing and rate. The experimental results show that compared to existing methods, the proposed method can achieve better sum rate performance with fewer pilots, effectively reducing pilot loss and reducing computational time complexity. And when the communication environment changes, the network also has good robustness. [ABSTRACT FROM AUTHOR]

In view of AI model production and training, the traditional script based physical server or cluster mode has problems such as training inference separated, insufficient resource utilization, difficult migration of computing environment, and lengthy training process. This paper proposed a platform architecture for deep learning model training, the architecture divided into four layers: data platform layer, computing platform layer, training suite layer, and management platform layer. Firstly, it proposed an integrated framework masks differences in network structures and optimized the graphs. Secondly, it researched an adaptive resource matching mechanism based on GPU state reduced communication costs. At the same time, it improved re- source utilization by providing a heuristic algorithm based label matching scheduling algorithm. Moreover, the establishment of tenant management and disaster recovery mechanisms ensured the safety and reliability of the system platform. Finally, it established the validation of the usability, safety, reliability, and scalability through the simulation platform. Through the construction of deep learning platform, it will accelerate the implementation of AI production and promote the prosperity and development of AI technology and ecology. [ABSTRACT FROM AUTHOR]

Single-view 3D reconstruction is an ill-posed problem. Due to the different modes of representation between the image and the 3D model, there are usually self-occlusion, low illumination, and multiple objects. Aiming at the ambiguity of the reconstructed model in the current 3D model reconstruction of a single image, this paper proposed a 3D point cloud model reconstruction method based on the guidance of prior information and multi-geometric angle constraints. Firstly, it obtained prior knowledge by pre-training the 3D point cloud encoder, and minimized the difference between the input image feature vector and the point cloud feature vector, so that the input image feature distribution approximated the point cloud feature distribution. Then, it used a differentiable projection module to project the three-dimensional point cloud representation of the image from different angles to a two-dimensional plane. Finally, it optimized the initial reconstructed point cloud by minimizing the diffe-rence between the projected image and the actual projected image in the dataset. The results of quantitative and qualitative comparison with other methods on ShapeNet and Pix3D datasets verify the effectiveness of the proposed algorithm. [ABSTRACT FROM AUTHOR]

The purpose of image stitching is to obtain a high-definition, seamless panoramic image. Existing methods rely on the accuracy of feature matching, which will misalign images and produce errors such as artifacts and distortions.This paper proposed a new unsupervised real scene image stitching network that can adapt to real scene stitching in the presence of moving targets, ensuring no loss of accuracy in the panorama. It included two networks of alignment and reconstruction, excluding the negative influence of moving targets and misleading regions on the transformation matrix through content branching, and optimizing image details by constraining the reconstruction process through edge branching to achieve high-definition and artifact-free stitching effects. The experimental results show that the method’s RMSE, PSNR, and SSIM reached 1.81, 26.56, and 0.85, respectively. The objective evaluation indexes are better than other classical algorithms overall, and the user research results also indicate that the method obtains higher definition of panoramic images. The method effectively accomplishes unsupervised image stitching in real scenes and can be generalized to stitching tasks in other scenes. [ABSTRACT FROM AUTHOR]

Fine-grained visual classification(FGVC) is extremely challenging due to the subtle inter-class differences and the large intra-class differences. In order to learn the embedded features of fine-grained images efficiently, this paper attempts to introduce the idea of knowledge distillation to FGVC, and proposed TRS-DeiT, which is equipped with the common advantages of CNN models and Transformer models simultaneously. Besides, a novel target regions selection module was proposed in TRS-DeiT to obtain the most discriminative regions. A contrastive loss function that measures the similarity of images was employed to distinguish the confusable classes in the task. Finally, the effectiveness of the proposed TRS-DeiT model was demonstrated through three datasets, CUB-200-2011, Stanford Cars and Stanford Dogs, which achieved the accuracy of 90.8%, 95.0% and 95.1% respectively. The experimental results shows that the proposed model outperforms the traditional models. Furthermore, the visualization results further illustrate that the attention learned by the proposed model mainly focuses on recognizing objects, thus contributes to fine-grained image classification tasks. [ABSTRACT FROM AUTHOR]

Existing lip-reading algorithms for Chinese sentences are inadequate at feature extraction and visual ambiguity resolution, which leads to low accuracy. Aiming at this problem, this paper proposed a lip-reading algorithm for Chinese sentences based on temporal convolutional network and 3D convolutional neural network（3DT-CHLipNet）.Firstly, it used a Temporal Convolutional Network to extract long-term features from lip dynamics sequences, which has a much larger receptive field than the Long Short Term Memory Network. Secondly, in order to minimize the visual ambiguity in Chinese lipreading, it adopted a Transformer model with the self-attention mechanism to capture the context information and improve the accuracy of sentence prediction. Finally, it introduced a temporal masking data augmentation strategy in the data preprocessing to further reduce the error rate of the algorithm. Comparison experiments on CMLR, the largest open-source Sequence-to-Sequence Chinese Mandarin Lip Reading dataset, show that the improvement in accuracy over representative lip reading algorithms for Chinese sentences ranges from 2.17%to 23.99%. [ABSTRACT FROM AUTHOR]

In the turbid underwater environment, due to the influence of multiple scattering and noise, the underwater imaging quality is greatly reduced. The existing significant target detection algorithms can't meet the task requirements of target detection accuracy. Therefore, this paper proposed an underwater salient target detection algorithm based on polarization imaging and salient region self-compensation. It divided the proposed algorithm into two stages. Firstly it used the method of estimating backscattered light based on polarization angle to remove the influence of backscattering, and introduced guided filter for underwater noise, and then removed the fuzzy effect caused by forward scattering through PSF function. Secondly it used an improved raster scanning algorithm combined with local feature descriptors to identify foreground superpixels, generated salient regions based on foreground superpixels to optimize and compensate the network, and then combined with DeepLabv3 + network to generate the final segmentation map. This paper conducted multiple sets of comparative experiments in a turbid underwater environment. The results show that the proposed restoration algorithm can effectively enhance the target, improve the contrast and clarity of the target, and the proposed salient target detection algorithm can accurately detect the underwater salient target and retain the detailed information. Combining the advantages of the polarization imaging model, this algorithm can better overcome the underwater complex optical imaging environment, and quickly and accurately detect significant underwater targets. [ABSTRACT FROM AUTHOR]

In dynamic gesture recognition methods based on deep learning, aiming at the problems of large network scale, simple dimension of extracted spatiotemporal features, and insufficient extraction ability of effective features, this paper proposed a deep network framework. Firstly, this paper applied a novel spatiotemporal convolution module based on multiscale information fusion in the time domain to improve the 3D residual network structure, greatly reduced the size of the network and obtained rich spatiotemporal receptive field characteristics. Then it introduced a spatiotemporal feature channel attention mechanism with global information synchronization, and used a few parameters to construct the global dependency between feature maps, and the module could obtain the key features of dynamic gestures more efficiently. The experimental results on the self-built gesture dataset SHC and the public gesture dataset SKIG show that the proposed gesture recognition method has fewer parameters and more powerful multiscale spatiotemporal feature extraction ability, and gains a higher recognition rate than the current mainstream algorithms. [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.)