Computer Engineering & Science

A parallel ambient noise data preprocessing algorithm based on heterogenous computing platform

WU Chao, WEI Qian, ZHOU Jun-wei, LI Hui-min, SUN Guang-zhong

2023, 45(10): 1711-1719. doi:

Abstract ( 279 )

PDF (1046KB) ( 320 ) 　　

Ambient noise seismology uses the ambient noise signals recorded by seismic stations to calculate the cross-correlation between stations, and thereby derive information about geological structures. In recent years, it has been widely used in fields such as Earth structure and oil and gas exploration. Seismic noise data processing often requires preprocessing calculations to reduce interference from instruments and seismic signals, which involves various signal processing calculations. As seismic stations are increasingly deployed in China, the continuous accumulation of seismic waveform files has greatly increased the time required for preprocessing calculations. To address the issue of computational time, a parallel seismic noise preprocessing algorithm has been proposed based on a graphics processing unit (GPU) heterogeneous computing platform. The parallel algorithm designed a parallel computing framework in three dimensions: stations, time, and segments. It implemented computational kernel functions for the calculation process in preprocessing and achieved adaptive processing of large batches of files through batch calculations. Experimental results show that the parallel preprocessing algorithm achieved an acceleration ratio of about 95 times, with good acceleration ratio and parallelism.

High-throughput parallel baseband processing algorithms based on GPUs for satellite communication

LI Rong-chun, ZHOU Xin, WANG Qing-lin, MEI Song-zhu

2023, 45(10): 1720-1730. doi:

Abstract ( 199 )

PDF (1165KB) ( 333 ) 　　

Satellite communication is widely used in fields such as encrypted communication and emergency communication. Its baseband processing algorithm is relatively complex and requires strong computing power support. Traditional FPGA and DSP platforms have a long development cycle, while GPU-based software-defined radio (SDR) solutions are convenient and have excellent performance. This paper proposes a group of baseband algorithms for satellite communication based on GPU, which realizes high-speed processing of the satellite communication downlink. Experimental results show that the GPU-based satellite communication link meets the lowest latency requirements, and the maximum baseband processing speed reaches 978Mbps.

A survey of precipitation nowcasting based on deep learning

MA Zhi-feng, ZHANG Hao, LIU Jie

2023, 45(10): 1731-1753. doi:

Abstract ( 736 )

PDF (1495KB) ( 733 ) 　　

Precipitation nowcasting refers to the high-resolution prediction of precipitation in the short term, which is an important but difficult task. In the context of deep learning, it is viewed as a radar echo map-based spatiotemporal sequence prediction problem. Precipitation prediction is a complex self-supervised task. Since the motion always changes significantly in both spatial and temporal dimensions, it is difficult for ordinary models to cope with complex nonlinear spatiotemporal transformations, resulting in blurred predictions. Therefore, how to further improve the model prediction performance and reduce ambiguity is a key focus of research in this field. Currently, the research on precipitation nowcasting is still in the early stage, and there is a lack of systematic classification and discussion about the existing research work. Therefore, it is necessary to conduct a comprehensive investigation in this field. This paper comprehensively summarizes and analyzes the relevant knowledge in the field of precipitation nowcasting from different dimensions, and gives future research directions. The specific contents are as follows: (1) The significance of precipitation nowcasting, and the advantages and disadvantages of traditional forecasting models are clarified. (2) The mathematical definition of the nowcasting problem is given. (3) Common predictive models are comprehensively summarized, analyzed. (4) Several open source radar datasets in different countries and regions are introduced, and download links are given. (5) The metrics used for prediction quality assessment are briefly introduced. (6) The different loss functions used in different models is discussed. (7) The research direction of precipitation nowcasting in the future is pointed out.

Implementation and optimization of HYB-based SpMV on the new-generation Sunway architecture

WANG Xin, PENG Jian

2023, 45(10): 1754-1762. doi:

Abstract ( 126 )

PDF (2235KB) ( 242 ) 　　

Sparse matrix vector multiplication (SpMV) is widely used in high performance computing. The parallelization of sparse matrix is more difficult than that of dense matrix because of the sparse and irregular distribution of non-zero elements. Therefore, the performance optimization of sparse matrix vector multiplication has always been the research focus in the field of high performance computing. A parallel SpMV algorithm and performance optimization scheme based on the HYB storage format of sparse matrices is designed for the new generation of domestic heterogeneous many-core processor SW26010P. Moreover, considering the difficulty of threshold selection in HYB storage format, a multi-iteration Otsu method is proposed to determine the threshold of HYB. The experimental results show that our design can achieve an average speedup of 23.36 and the best speedup of 34.85, compared with the sequential method on the Main Processing Element (MPE) of SW26010P.

Research on the analysis and optimization of power switch noise coupling interference on high-speed signals

GONG Wei, LI Yan, XU Si-qiang, QI Hong-yu

2023, 45(10): 1763-1769. doi:

Abstract ( 134 )

PDF (1881KB) ( 223 ) 　　

In electronic systems, the problem faced by high-speed circuit design ultimately lies in noise. In the actual high-speed circuit design and production process, there are many factors that can cause problems with signal integrity and power integrity, such as deviations between simulation models and actual PCBs, unreasonable designs, and errors generated during PCB board manufacturing. To address these issues, we conduct research on the analysis and optimization of power switch noise coupling interference on high-speed signals. We first introduce the theory of signal integrity and power integrity, and then conduct theoretical analysis, oscilloscope testing analysis, Cadence simulation analysis, and optimization verification on a case where power switch noise coupling interference affects high-speed signals in a motherboard design process. This paper provides a systematic analysis and solution for similar problems.

An attribute-based dynamic mandatory access control mechanism for operating system

DING Yan, WANG Peng, WANG Chuang, LI Zhi-peng, SONG Lian-tao, FENG Liao-liao

2023, 45(10): 1770-1778. doi:

Abstract ( 119 )

PDF (1213KB) ( 277 ) 　　

Mandatory access control (MAC) for operating system (OS) brings strong security guarantee for the system because it runs at high privilege level. However, the classical OS MAC only supports static security policies. When the security requirements change, the security policies must be reconfigured and reloaded. Therefore, it is difficult to meet the requirements of dynamic regulation of access permissions in scenarios such as high-sensitivity application state transition, cloud native dynamic scheduling, and BYOD. Attributes-based access control has strong extensibility, flexibility and expression ability, which provides a solution to improve the dynamic and flexibility of the security policy of MAC in OS. In this paper, the theoretical model and system architecture model of attributes-based dynamic mandatory access control for operating systems are proposed. Then, the prototype system is designed and implemented by combining with the classic MAC mechanism of Linux, and the feasibility of the model is verified. Finally, in view of the possible performance impact of the introduction of attribute factors, the optimization research of access control is carried out from two aspects of time and space.

An attribute-based encryption scheme supporting complex access policies

XU Cheng-zhou, LI Lu, ZHANG Wen-tao

2023, 45(10): 1779-1788. doi:

Abstract ( 134 )

PDF (851KB) ( 209 ) 　　

Aiming at the access structure of attribute-based encryption, this paper proposes an attribute-based encryption scheme that supports complex access policies. The scheme uses reduced ordered binary decision diagrams (ROBDDs) as the access structure, where a user's attribute set corresponds to a path in the ROBDD. The ROBDD can not only represent any Boolean function about attri- butes, but also reduce valid paths by simplifying nodes in the access structure, thereby preventing interference from irrelevant attributes and reducing the computational cost in the encryption phase. By integrating effective path feature values into Boolean functions, the ciphertext does not need to store multiple effective path feature values in complex access policies, reducing the storage cost of the ciphertext. The scheme outsources attribute authentication computation to the decryption server, reducing the local computation cost of users in the decryption phase, and uses group element exponentiation instead of bilinear pairing to reduce the computational cost of the scheme. The security model proves that the scheme is IND-CPA secure, and performance analysis and experimental simulation show that the proposed scheme has lower computational and storage costs.

Text steganography based on generative adversarial networks and multi-head attention

HUANG Yao, PAN Li-li, XIONG Si-yu, JIANG Xiang-hui, MA Jun-yong

2023, 45(10): 1789-1796. doi:

Abstract ( 178 )

PDF (875KB) ( 236 ) 　　

With the development of deep learning, steganography based on text generation has made significant break-throughs. Existing text-based steganography methods suffer from exposure bias, where the input during training comes from real sample labels, while the input during prediction comes from the output predicted in the previous time step. This difference in input samples between training and prediction leads to error accumulation, resulting in a large distribution difference between generated samples and real samples. To address this problem, this paper proposes a text steganography model called TS-GANMA based on generative adversarial networks and multi-head attention. First, a text generator is trained using a generative adversarial network, and multi-head attention mechanisms are used to extract multi-head attention scores to participate in the reward calculation of the reward module, obtaining feedback information more suitable for the generator. Then, the generator and discriminator are trained in an adversarial manner, which can solve the exposure bias problem and optimize the text generation model. Finally, the conditional probability distribution output by the text generation model is encoded to embed secret information. The experimental results demonstrate that the steganography method based on TS-GANMA has a much lower perplexity than the methods based on LSTM-vlc and ADG at the same embedding rate. This is because the steganographic text generated by the TS-GANMA model fits the statistical distribution of the real text better, and can generate higher quality steganographic text.

A trajectory data differential privacy protection scheme that combines contrast supervision and sorting tree

WANG Hui, CHEN Yu, SHEN Zi-hao, LIU Pei-qian

2023, 45(10): 1797-1805. doi:

Abstract ( 103 )

PDF (943KB) ( 265 ) 　　

With the popularization of various devices that provide location positioning services, while users enjoy the convenience brought by these devices, it also raises the issue of location privacy leakage. To address this problem, a trajectory data differential privacy protection scheme (SDTS) that combines contrast supervision and sorting tree is proposed. First, the supervised learning model is used to preprocess the trajectory data, and the loss function in the model is used to calculate the trajectory similarity and obtain the result. Second, a binary search tree structure is used to store the trajectory data, improv- ing the efficiency of trajectory queries. Finally, differential privacy technology and an equal privacy budget allocation method are used to add noise to the statistical values of moving users in the sorted tree nodes, protecting sensitive information stored in the nodes and ensuring data privacy security while improving data usability. Experimental results show that this scheme effectively protects users data privacy security and ensures the usability of trajectory data.

Software design of acoustic scanning image defect detection based on machine vision

ZHAO Yue, XIAO Meng-yan, QIU Bao-jun, LUO Jun, WANG Xiao-qiang, LUO Dao-jun

2023, 45(10): 1806-1813. doi:

Abstract ( 121 )

PDF (1891KB) ( 214 ) 　　

Integrated circuits are an important part of electronic products, and their quality control and fault analysis are prerequisites for the long-term operation of electronic products. Scanning Acoustic Microscope (SAM), as a non-destructive defect detection method, has been widely used in imaging detection and internal defect identification of integrated circuits. In response to the intelligent demand for acoustic scanning image defect detection and the requirements for real-time and accurate detection, this paper develops a software for integrated circuit acoustic scanning image defect detection based on machine vision, providing integrated functions for image processing and image detection. The algorithm framework of this software combines deep learning technology, traditional image processing technology using OpenCV, and JavaScript interface design technology, allowing it to manage various types of integrated circuit data and analyze, process, and determine defects in scanning acoustic images.

A video human behavior recognition method based on improved 3D ResNet

NIU Wei-hua, ZHAI Rui-bing

2023, 45(10): 1814-1821. doi:

Abstract ( 168 )

PDF (745KB) ( 202 ) 　　

Aiming at the temporal characteristics of human behavior in videos, a video human beha- vior recognition method is proposed that combines asymmetric convolution and CBR modules. This method uses 3D ResNet-50 as the backbone network. First, the larger convolutions in the network are changed to the concatenation of two asymmetric 3D convolutions, which deepens the local key feature extraction of the convolution layer in the horizontal and vertical directions. Secondly, CBR module is added to improve the number of network layers. The network extracts multi-angle features of images and time series from continuous video frame sequences, classifies them according to the feature data, and finally outputs the recognition results. Extensive experimental results on the benchmark dataset UCF101 show that the Top1 and Top5 accuracy of the proposed method are improved by 4.03% and 4.99%, respectively, compared with the original 3D ResNet network, and the recognition accuracy of this method is also better than other mainstream methods.

A grasp pose estimation method combining semantic instance reconstruction

HAN Hui-yan, WANG Wen-jun, HAN Xie, KUANG Li-qun, XUE Hong-xin,

2023, 45(10): 1822-1829. doi:

Abstract ( 124 )

PDF (1025KB) ( 201 ) 　　

To solve the problem that it is difficult to distinguish multiple adjacent objects and the accuracy of high-dimensional pose learning is poor, a pose estimation method combining on semantic instance reconstruction is proposed. The semantic instance reconstruction branch is added to complete implicit 3D reconstruction of the foreground, and the center coordinate of each foreground point belongs to the instance is predicted by the voting method to distinguish adjacent objects. A pose dimensionality reduction learning method is proposed. Two orthogonal unit vectors are used to decompose the three- dimensional rotation matrix to improve the accuracy of pose learning. A semantic instance reconstruction grasping network (SIRGN) is proposed, and the training is completed on VGN simulation grasping dataset. The experimental results show that the grasping success rate of SIRGN in Packed and Pile environment is 89.5% and 78.1% respectively, and it has good applicability in real environment.

A low-light image enhancement algorithm based on multi-scale depthwise separable convolution

CHEN Qing-jiang, GU Yuan

2023, 45(10): 1830-1837. doi:

Abstract ( 132 )

PDF (1133KB) ( 272 ) 　　

To address the issues of color distortion and low contrast in low-light images, and severe detail loss and excessive parameters of existing enhancement algorithms, a low-light image enhancement algorithm based on multi-scale depthwise separable convolution is proposed. Firstly, a multi-scale hybrid dilated convolution module is designed to expand the receptive field while addressing grid effects. Secondly, a multi-scale feature extraction module is designed to extract feature information at different scales. Finally, the two modules are used to fully integrate low-level spatial information with high-level semantic information for different-sized feature maps to obtain the final output. The use of depthwise separable convolution instead of standard convolution greatly reduces the network parameter count and computational cost. Experimental results show that the proposed algorithm effectively improves the brightness and contrast of images, reduces the number of model parameters, and restores image texture details and color well.

A dense multi-face detection algorithm based on YOLOv5s

DONG Zi-ping, CHEN Shi-guo, LIAO Guo-qing

2023, 45(10): 1838-1846. doi:

Abstract ( 194 )

PDF (1922KB) ( 358 ) 　　

To address the problem of missed detection in dense scenes and low detection rate for small-scale faces, an improved multi-face detection algorithm based on YOLOv5s, named IYOLOv5s-MF, is proposed. First, the feature texture transfer (FTT) module is introduced into the feature fusion part to obtain more feature representations for small-scale faces. Then, the positive and negative sample sampling strategy is improved by increasing the number of effective positive samples to enhance the model's generalization ability. Finally, Focal-EIoU is adopted as the localization loss function to accele- rate model convergence and improve face detection accuracy. Experimental results on the WIDER FACE dataset show that compared with other comparison algorithms, IYOLOv5s-MF has higher face detection accuracy and good real-time performance.

A question generation model with multi-stage temporal and semantic information enhancement

ZHOU Ju-xiang, ZHOU Ming-tao, GAN Jian-hou, XU Jian

2023, 45(10): 1847-1857. doi:

Abstract ( 124 )

PDF (845KB) ( 179 ) 　　

To address the problems of multi-stage encoding of the encoder of the graph-to-sequence question generation model and the easy loss of rich sequence information and semantic structure information in the paragraphs during decoding, this paper designs a model based on multi-stage timing and semantic information enhancement(MS-SIE). The model first fuses the semantic information of the passages encoded at different stages of the encoder and inputs them to the recurrent neural network for encoding. Then, an iterative graph neural network is introduced in the decoding stage to combine the encoded paragraph information with the rich semantic structure information hidden in the previously generated text questions in the decoding stage. Finally, a recurrent neural network based on an attention mechanism is used to generate the questions. The results show that the model proposed significantly outperforms the existing sequence-to-sequence model and graph-to-sequence model in both automatic evaluation metrics and manual evaluation metrics.

Aspect-level sentiment analysis of graph attention network fused with graph walk information

YANG Chun-xia, GUI Qiang, MA Wen-wen, XU Ben,

2023, 45(10): 1858-1865. doi:

Abstract ( 109 )

PDF (799KB) ( 201 ) 　　

In the aspect-level sentiment analysis task, the attention mechanism is commonly used to obtain the weight information of words, ignoring the effect of syntactic structure on extracting the importance of different words in sentences. In addition, there is a problem of confusion between aspect words and sentiment words in multi-aspect word sentences. How to effectively pay attention to the context part related to the emotional polarity of the target aspect words is often ignored. A graph attention neural network model (GW-GAT) that integrates graph walk information is proposed. Graph walk is performed on the grammar graph to obtain the word weight coefficient of the sentence. The graph attention network is used to combine the word node weight and the weight between the nodes to highlight the context part that plays an important role in the word sen-timent polarity in the target aspect. Finally, the sentiment polarity is obtained through fully connected and softmax layers. The experimental results on the SemEval2014 task and the Twitter dataset show that the GW-GAT model outperforms the baseline model and obtains better experimental results.

Aspect-level sentiment classification based on dual attention fusion knowledge

ZHANG Qian-kun, HAN Hu, HAO Jun

2023, 45(10): 1866-1873. doi:

Abstract ( 127 )

PDF (642KB) ( 232 ) 　　

Aspect-level sentiment classification aims to discriminate the sentiment polarity of a specific aspect in a sentence. Although attention-based recurrent neural network models perform well among existing solutions, they are not ideal for processing "semantically ambiguous" sentences that are short and contain many neologisms and polysemous words. Therefore, this paper proposes a neural network model based on knowledge graph and attention mechanism. The basic idea is to use a knowledge base to obtain a relevant concept set of aspect words and integrate external information to enhance the semantic representation of the text. Firstly, the output of bidirectional long short-term memory network is combined with self-attention mechanism to generate context representation. Then, the upper and lower context representations are combined to use dual attention to obtain external knowledge from the knowledge graph and obtain knowledge vectors related to aspect words. Finally, the two parts of content are input together into a fully connected network to calculate the aspect-level sentiment tendency. Experimental results show that compared with other models, the proposed model significantly improves classification performance.

A folk songs fast classification algorithm DUPSO-DSVM based on distance sorting

Lv Xiao-jiao, ZHANG Yu-mei, YANG Hong-hong, WU Xiao-jun,

2023, 45(10): 1874-1833. doi:

Abstract ( 90 )

PDF (3150KB) ( 179 ) 　　

In the context of the rapid development of network information, the demand of different music lovers for music information retrieval is also increasing, and music classification has become an important research subject. This paper proposes a fast classification method of DUPSO-DSVM folk songs, which combines dissipative uniform particle swarm optimization (DUPSO) with distance sorted SVM (DSVM). This method uses DUPSO algorithm to optimize the penalty coefficient C and kernel function parameter g of SVM, and uses DSVM algorithm to optimize the parameter optimization time of DUPSO algorithm. The experimental results show that, DUPSO-SVM algorithm has a classification accuracy of 84%. After using DUPSO-DSVM algorithm, the training time of the algorithm only accounts for 26.26% of the unused DUPSO-DSVM algorithm, but it still maintains a high classification accuracy.

A forest fire image early warning detection method based on probabilistic two-stage CenterNet2

LI Bao-min, WANG Xiao-peng, SUN Qian-rong, ZHANG Jun-ping

2023, 45(10): 1884-1890. doi:

Abstract ( 91 )

PDF (1416KB) ( 190 ) 　　

Timely warning of forest fire plays a crucial role in forest protection. Due to the complex background of forest fireworks and many interference factors, the detection accuracy and efficiency are affected. Therefore, a forest fire image detection method based on CenterNet2 is proposed. The lightweight backbone network VoVNetV2 combined with asymmetric convolution kernel is used to improve the feature extraction ability and detection speed. Meanwhile, an attention mechanism eSE (Effective Squeeze and Extraction) is introduced into the weighted bidirectional feature pyramid network for feature fusion, so as to improve the accuracy of small target detection. Then, SIoU loss function is used to improve the effect of target box regression. The simulation results show that the method can accurately detect forest fire in real time, and the false rate is low.

Deep input-aware factorization machine based on Setwise ranking

LIU Tong, ZHOU Ning-ning

2023, 45(10): 1891-1900. doi:

Abstract ( 85 )

PDF (1043KB) ( 166 ) 　　

SetRank is a novel Setwise Bayesian collaborative ranking model that can model implicit feedback data more closely to real-world scenarios. However, SetRank only considers collaborative information and lacks effective use of content information. In order to solve the above problems, a factorization machine model based on Setwise ranking, named SRFMs, is proposed. Drawing on the Setwise ranking proposed in SetRank to solve the problem of missing negative samples in implicit feedback, a factorization machine is chosen as a predictor to model content information, and model user preferences from the perspective of optimal item ranking. Furthermore, to improve the fixed feature representation and the lack of higher-order feature interactions in standard FM, inspired by IFM, SR-DIFM model is constructed to incorporate the SRFMs by combining input-aware networks with neural networks. Experimental results on two real-world datasets demonstrate that the proposed model outperforms the state-of-art model in terms of evaluation metrics including HR, NDCG and mAP, and can improve the accuracy of recommendations which make better use of user and item content information while solving the recommendation problem under implicit feedback.

Current Issue

Author center

Review center

Online journal