• Volume 32,Issue 10,2023 Table of Contents
    Select All
    Display Type: |
    • Landmark Localization in Medical Images Based on Uncertainty Estimation

      2023, 32(10):1-9. DOI: 10.15888/j.cnki.csa.009247

      Abstract (837) HTML (1046) PDF 1.98 M (1957) Comment (0) Favorites

      Abstract:Heatmap-based methods are mainstream for landmark localization in medical images. However, current heatmap-based methods almost exclusively employ predetermined heatmaps as labels, which cannot fully represent the real location distributions of landmarks to limit performance. Therefore, this study proposes a landmark localization algorithm in medical images based on uncertainty estimation to simultaneously predict the landmarks and their location distributions. The model adopts multi-branch dilated convolution to extract multi-scale context information and employs a self-attention mechanism to enhance important features, thus improving the landmark detection ability while predicting the distributions. Experiments on public datasets show that the proposed method improves the overall landmark detection performance and performs better on most metrics. Additionally, the predicted distributions are consistent with the real annotation distributions.

    • Selection and Prediction of Multiple Influencing Factors of Cotton Price Based on XGBoost and TCN-Attention

      2023, 32(10):10-21. DOI: 10.15888/j.cnki.csa.009262

      Abstract (724) HTML (1347) PDF 2.88 M (1858) Comment (0) Favorites

      Abstract:Cotton price is complex and changeable due to many factors, and the prediction accuracy of cotton price can be improved by selecting appropriate data features and prediction models. In this study, the daily spot price data of cotton are taken as the research target, and nine influencing factors in four aspects of supply and demand, international market, macroeconomy, and industrial chain are collected as features. The extreme gradient boosting (XGBoost) algorithm is used to evaluate and screen the influencing factors of cotton price, and five of them are selected. This study adopts the time convolution network (TCN) with an attention mechanism (Attention), namely TCN-Attention, TCN, long short-term memory (LSTM), gate recurrent unit (GRU), and other models to predict cotton price. Through ablation experiments and comparative experiments, the results show that: (1) After XGBoost feature screening, the mean absolute error (MAE) and root mean square error (RMSE) of TCN-Attention price prediction are 41.47 and 58.76, which are 77.57% and 76.49% lower than those before screening; (2) compared with TCN, LSTM, and GRU, the TCN-Attention model proposed in this study has more accurate prediction results. MAE and RMSE are reduced by more than 50%, and the running time is shortened by 60% compared with LSTM and GRU.

    • Hand Pose Estimation Based on Multi-view Learning Strategy

      2023, 32(10):22-33. DOI: 10.15888/j.cnki.csa.009291

      Abstract (724) HTML (1290) PDF 14.69 M (1590) Comment (0) Favorites

      Abstract:Hand pose estimation plays an important role in human-computer interaction, hand function assessment, virtual reality, and augmented reality. Therefore, a new hand pose estimation method is proposed to handle the relatively small proportion of hand region in most images and the occlusion problem of single-view keypoint detection algorithms. The proposed method first extracts the hand target region by using a semantic segmentation model which introduces the Bayesian convolutional neural networks. According to the hand localization result, the proposed method adopts a new model based on the attention mechanism and cascade guidance strategy to obtain accurate 2D hand keypoint detection results. Then, the proposed method uses a deep network based on a stereo vision algorithm to calculate the depth information of the keypoints, and the view self-learning function is provided in depth estimation. The algorithm uses triangulation as the foundation, and the RANSAC algorithm is used to correct the measurement results. Finally, the 3D hand keypoint detection results can be optimized by using multi-task learning and reprojection training, and the 3D pose of the hand keypoints can be obtained. Experimental results show that compared with some representative hand region detection algorithms, the proposed method has a significant improvement in the average detection precision and running time for hand regions. In addition, in terms of the end-point-error mean (EPE_mean) and the area under PCK curve (AUC) of different pose estimation methods, it can be seen that the keypoint detection performance of the proposed method is better. Thus, a better hand pose estimation result can be obtained.

    • HyperLedger Fabric Performance Optimization Based on Key-value State Sharding

      2023, 32(10):34-44. DOI: 10.15888/j.cnki.csa.009253

      Abstract (684) HTML (1339) PDF 2.01 M (1665) Comment (0) Favorites

      Abstract:HyperLedger Fabric is an open-source consortium blockchain that has received a lot of attention. Since the existing blockchain sharding method is not suitable for the three-stage transaction model of Fabric, and there is the problem of hotspot access caused by coarse sharding granularity, a fine-grained key-value state sharding method based on Fabric is proposed. First of all, the cross-shard transaction processing of Fabric under key-value state sharding is designed in detail, and a cross-shard sequence node and a two-stage submission process are introduced to efficiently ensure the consistency and atomicity of cross-shard transactions. Then, in view of the problem that fine-grained sharding may lead to an increase in the probability of transaction cross-shards and thus affect performance, a heuristic transaction proposal routing table is proposed to reduce the cross-shard data read requests of transactions in the pre-execution stage and lower the consumption of computing and network resources. Finally, the improved sharding scheme and performance test are realized on the Fabric simulation system. The experimental results show that on the basis of improving the performance of Fabric, this method effectively solves the hotspot access problem and the performance degradation problem under the high proportion of cross-shard transactions.

    • Recognition of Cuttings Images Based on Improved BiSeNetV1 Real-time Model

      2023, 32(10):45-53. DOI: 10.15888/j.cnki.csa.009245

      Abstract (687) HTML (1029) PDF 1.91 M (1704) Comment (0) Favorites

      Abstract:In the field of image segmentation and identification, the existing deep learning methods mostly perform tasks by high-precision semantic segmentation methods, which lead to a slow network inference speed, large amount of calculation, and difficult actual application. A real-time network model with better performance, namely BiSeNetV1 is used, and the extended spatial path convolution structure, spatial pyramid attention mechanism (SPARM), simplified iterative attention feature fusion (S-iAFF) module, and other optimization strategies are applied. As a result, a real-time BiSeNet_SPARM_S-iAFF network is designed for rock debris image segmentation. The extended spatial path convolution structure can obtain more abundant spatial features of rock debris images. The context path uses the optimized SPARM to further refine high-level semantic feature extraction. Finally, S-iAFF is used to enhance the fusion degree between low-level spatial and high-level semantic features in the feature fusion stage. The experimental results indicate that the mean intersection over union (mIoU) of the BiSeNet_SPARM_S-iAFF network on the RockCuttings_Oil dataset is 64.91%, which is 2.68% higher than that of the BiSeNetV1 network, and the precision of the improved network is close to that of the most high-precision semantic segmentation methods, while the number of parameters is greatly reduced, and the inference speed is significantly improved.

    • Stock Index Prediction with Text Generative Language Model

      2023, 32(10):54-64. DOI: 10.15888/j.cnki.csa.009266

      Abstract (664) HTML (1155) PDF 3.03 M (1708) Comment (0) Favorites

      Abstract:Stock index prediction is an important topic in the field of finance. With the development of computing power and technologies, there are opportunities to improve the performance of stock index prediction by identifying and quantifying valuable information from online news. In order to extend the econometric literature on stock index prediction frameworks to high-dimensional textual data, a stock index prediction framework based on generative language models is proposed. The prediction framework can be divided into two steps. First, a supervised generative language model is used to filter out noisy words quickly and aggregate the remaining text into a news index that can fully explain stock index changes. Second, the news index and historical stock index data are jointly used as independent variables of the time-varying parameter predictive model to predict future stock index values. The framework not only enriches the influencing factors of stock index prediction but also reveals the time-varying dynamic relationship between these factors and stock index values. Empirical research demonstrates the explanatory and out-of-sample predictive power of the proposed prediction framework. Among the six industrial stock indices predicted, the mean square error obtained by the proposed prediction framework is generally lower than that by traditional time series and machine learning methods. Compared with the time-varying parameter predictive model and long short-term memory model that do not consider news information, the proposed prediction framework also exhibits better predictive performance.

    • Property Graph Storage System Based on Persistence Memory

      2023, 32(10):65-74. DOI: 10.15888/j.cnki.csa.009274

      Abstract (507) HTML (954) PDF 1.71 M (1482) Comment (0) Favorites

      Abstract:A property graph is a popular graph data model that has been widely used in various graph systems. However, when coming to graph analysis workloads, graph database systems for transactional workloads encounter challenges in terms of high latency. Traditional graph analysis systems are geared towards simple graph models and have limited transactional workload support of graphs. Therefore, there is a growing demand for a graph storage system that can efficiently handle both graph analysis tasks and transactional workloads on property graphs. The emergence of persistent memory provides people with an opportunity to redesign graph storage systems to fully leverage the advantages of this device. To this end, this study proposes TAG, a persistent memory-based property graph storage system. TAG adopts a novel hybrid architecture for graph storage to fully utilize the advantages of persistent memory and main memory. Secondly, by combining topology and index into one, TAG embeds the graph topology into the system index to accelerate queries on the graph topology. Finally, by organizing the graph's property data based on labels, TAG further optimizes access to graph properties. Experimental results show that TAG is significantly better than other graph database systems and has comparable performance to graph analysis systems.

    • Use Case Identification Based on UML Activity Diagram

      2023, 32(10):75-84. DOI: 10.15888/j.cnki.csa.009265

      Abstract (606) HTML (774) PDF 5.82 M (1547) Comment (0) Favorites

      Abstract:In the object-oriented software development process, use case diagrams of unified modeling language (UML) are applied to capture the user requirements. The traditional method of describing use cases is generally based on the developer's own experience to obtain use cases from the requirements manually. However, how to automatically generate accurate use cases is still a problem to be solved. This study proposes a method to generate use cases semi-automatically by using UML activity diagrams. Firstly, the study specifies the use case diagram and activity diagram by introducing a formal model, the unified structure. Secondly, it gives an algorithm for decomposing the activity diagram and then generates the event flow of the corresponding use cases, which is based on the dependency chain obtained from the decomposed activity diagram, to obtain the use case model. Finally, the case is demonstrated by the developed prototype CASE tool and the feasibility of the proposed method is verified.

    • Music Genre Classification Based on Spectrogram Enhancement and CNNBLS

      2023, 32(10):85-95. DOI: 10.15888/j.cnki.csa.009272

      Abstract (597) HTML (2278) PDF 4.34 M (2040) Comment (0) Favorites

      Abstract:For the problems of weak music feature mining, complex deep learning classification models, and long training time, a music genre classification model based on spectrogram enhancement and convolutional neural network-based broad learning system (CNNBLS) is designed. This model first enhances the Mel spectrogram by randomly masking part of frequency channels in SpecAugment and then uses the cut Mel spectrogram as the input of CNNBLS. At the same time, exponential linear unit functions (ELUs) are fused into the convolutional layer of CNNBLS to enhance its classification accuracy. Compared to other machine learning network frameworks, CNNBLS can achieve higher classification accuracy with less training time. In addition, CNNBLS can quickly learn incremental data. The experimental results show that the non-incremental model of CNNBLS can achieve a classification accuracy of 90.06% after training 400 pieces of music data, while the incremental model of Incremental-CNNBLS can achieve a classification accuracy of 91.53% after adding 400 pieces of training data.

    • Hybrid-attention-based Lightweight Hemiplegic Gait Assessment System

      2023, 32(10):96-105. DOI: 10.15888/j.cnki.csa.009260

      Abstract (679) HTML (970) PDF 3.23 M (1640) Comment (0) Favorites

      Abstract:Stroke patients often exhibit hemiparetic gait, and visual gait analysis can be applied to detect such changes. However, publicly available pathological gait datasets are small in scale and lack detailed grading of hemiplegia severity. Furthermore, state-of-the-art deep learning algorithms for gait analysis usually have a high need for parameter size and computational complexity, leading to low performance on small-scale pathological gait datasets. To address these challenges, this study designs a lightweight hemiplegic gait recognition system. The system utilizes an attention-based lightweight convolutional neural network (CNN) to access hemiplegic gait performance. By linear splicing grouped convolution at different scales, high-efficiency features can be obtained at a low cost. Additionally, a multidimensional hybrid lightweight attention module is introduced to assist CNN in focusing on distinctive features in both spatial and channel dimensions, achieving a good balance between system effectiveness and lightweight design. Moreover, a hemiplegic simulation gait dataset is constructed, specifically for hemiplegic gait recognition to support model training and testing. The results demonstrate that the proposed network that uses only 1/53 parameters of VGG-19 improves the accuracy of gait recognition to 96.91%, which is higher than that of pre-train VGG-19. Compared with other lightweight SOTA methods, it also has the advantage of accuracy. The system has low development costs and can be deployed on mobile devices. It supports real-time detection, providing a feasible solution for home-based pathological gait analysis.

    • GUI-based Visualization System for Plasma Control Simulation on EAST

      2023, 32(10):106-114. DOI: 10.15888/j.cnki.csa.009241

      Abstract (644) HTML (1034) PDF 2.40 M (1696) Comment (0) Favorites

      Abstract:Simulation suites for plasma control on EAST (SPACE), developed based on the open-source language Python, is a function library for plasma control simulation of magnetic confinement fusion on EAST tokamak. The main feature of SPACE is to analyze, design, predict, and simulate the tokamak plasma control on the basis of the tokamak device model, plasma physical model, and control system model by using computer numerical simulation technology. In order to meet the requirements of the visual operation of SPACE functional modules, a visual operation system of plasma control simulation suitable for EAST superconducting tokamak is developed by using Python and PySide2. The system enables researchers to carry out related operations of plasma control simulation by means of graphic interaction and significantly improves the efficiency of plasma control simulation.

    • User Interface and Parameter Management of Plasma Control System

      2023, 32(10):115-122. DOI: 10.15888/j.cnki.csa.009242

      Abstract (507) HTML (832) PDF 3.34 M (1477) Comment (0) Favorites

      Abstract:Plasma control system is one of the critical systems in fusion experiments, responsible for real-time feedback control of various plasma parameters. The existing independently developed plasma control system adopts a component-based model, with its core control function implemented by executing various plasma control algorithms through algorithm components. It requires preset parameters of the graphical interface editing algorithm and will manage them. Therefore, the graphical interface is implemented by using PyQt5, and a parameter configuration component is developed to process parameter storage and retrieval. Data transmission between them is achieved through publish/subscribe messaging mechanism. Extensible markup language is used to define the preset algorithm parameter information, so as to decouple the user interface and control algorithm. The parameter data structure is uniformly defined through interface description language, and MySQL database is designed to store historical parameter data. Data communication is completed based on publish/subscribe messaging mechanism.

    • Augmented Reality Method Combining Object Detection and Spatial Projection

      2023, 32(10):123-131. DOI: 10.15888/j.cnki.csa.009271

      Abstract (507) HTML (1023) PDF 2.02 M (1454) Comment (0) Favorites

      Abstract:To register and track fixed objects, the common methods are using prefabricated markers, or using professional AR devices with integrated depth cameras and other accessories, whose costs are high. To address the defects of existing methods, a simple cooperative hybrid tracking and registration technology that integrates object detection and spatial projection algorithm is proposed. Firstly, the object type is obtained by the deep learning algorithm for object detection, and then the specific object ID is determined by the spatial projection algorithm using position and posture information obtained from sensors, which improves the matching degree and accuracy of the virtual information superimposed on the real scene. Based on this algorithm, an AR application for smart IoT infrastructure maintenance is realized and experiments are conducted on objects such as light poles and trashcans. The experimental results show that this method can run on ordinary smartphones and AR glasses, achieving the expected results, avoiding the need for prefabricated markers, and reducing the requirement for hardware resources.

    • SoC Design of Chinese-braille Translation Based on Cortex-M3

      2023, 32(10):132-139. DOI: 10.15888/j.cnki.csa.009256

      Abstract (599) HTML (940) PDF 2.90 M (1464) Comment (0) Favorites

      Abstract:Chinese-braille translation refers to the process of automatically translating a Chinese text into corresponding braille data. In an embedded environment, the speed of Chinese-Braille translation is relatively slow, and it is difficult to meet the real-time requirements in complex environments. Therefore, an IP core for dedicated Chinese-braille translation is designed, which can achieve accurate braille data by implementing a reverse maximum matching segmentation algorithm and Chinese-braille conversion. In order to verify the rationality of the design, the SoC is constructed with Cortex-M3 as the microprocessor, which is equipped with serial ports, LCD drivers, and an IP core for Chinese-braille translation. Functional verification and performance testing are carried out by using FPGA experimental platform. The test results show that the SoC can accurately perform Chinese-braille translation at a speed of 5 079.37 B/s.

    • Acoustic Scene Classification Based on Hierarchical Information Fusion

      2023, 32(10):140-146. DOI: 10.15888/j.cnki.csa.009258

      Abstract (525) HTML (811) PDF 1.08 M (1547) Comment (0) Favorites

      Abstract:Acoustical scene classification technology plays an important role in daily life by analyzing its recording environment through the audio recorded in public areas. Different from the traditional classification problem in which there is no relationship between classes, there is an implicit hierarchical structure relationship between the classes of acoustic scene classification (parent class and subclass). For example, the parent class of the airport and shopping mall is indoor. However, the existing methods do not consider this characteristic of acoustic scene classification task and ignore the dependency relationship between the parent class and the subclass. Therefore, an acoustic scene classification method is proposed, which is based on hierarchical information fusion by using the hierarchical structure relationship between acoustic scene classes. In this method, two separate classifiers are designed to classify the parent class and the subclass respectively. The information of the parent class is fused in the process of the subclass classification, and the hierarchical dependency loss is designed to punish the predicted mismatch between the parent class and the subclass. The experimental results on TAU urban acoustic scenes 2020 mobile development dataset show that the method based on hierarchical information fusion effectively improves the performance of the acoustic scene classification model with an increase of 1.1% in classification accuracy.

    • Border Peeling Clustering Based on Shared Nearest Neighbors and Optimized Association Strategy

      2023, 32(10):147-156. DOI: 10.15888/j.cnki.csa.009263

      Abstract (656) HTML (876) PDF 3.32 M (1237) Comment (0) Favorites

      Abstract:The border peeling (BP) clustering algorithm is a density-based clustering algorithm. It gradually peels up border points to reveal the potential cores of clusters and has been proven to be an effective clustering algorithm. However, the BP algorithm has some limitations. On the one hand, the local density of data points only considers the distance characteristics, which can lead to the unreasonable determination of border points. On the other hand, the association strategy of the BP algorithm is prone to misjudge outliers and can generate associated errors when border points are allocated. Hence, this study proposes a BP clustering algorithm based on shared nearest neighbors and optimized association strategy (SOBP). The algorithm employs a local density function based on shared nearest neighbors to better explore the similarity between data points. Meanwhile, the association strategy of the BP clustering algorithm is optimized so that in each iteration, border points are no longer associated with only one non-border point. Furthermore, a double association criterion between border points and non-border points as well as between border points peeled up is utilized. Tests on several datasets show that the proposed algorithm outperforms six other classical algorithms in terms of evaluation indexes.

    • Improved YOLOv7 Algorithm for Traffic Sign Detection

      2023, 32(10):157-165. DOI: 10.15888/j.cnki.csa.009227

      Abstract (845) HTML (1955) PDF 3.07 M (2443) Comment (0) Favorites

      Abstract:The rapid development of autonomous driving technology has led to increasing requirements for traffic sign detection technologies. In order to solve the problems of false detection and missed detection of the YOLOv7 algorithm in identifying small targets, this study proposes a traffic sign detection model based on an attention mechanism, namely YOLOv7-PC. Firstly, a K-means++ clustering algorithm is used to cluster the traffic sign dataset to obtain anchor boxes suitable for detecting traffic signs. Secondly, the coordinate attention mechanism is introduced into the YOLOv7 backbone feature extraction network, and the horizontal and vertical information of traffic signs are embedded into the channel so that the generated feature information has the coordinate information of traffic signs, and the extraction of effective features is enhanced. Finally, the atrous spatial pyramid pooling is introduced in the enhanced feature extraction network to capture multi-scale context information of traffic signs, which ensures the resolution of small targets of traffic signs and expand the receptive field of the convolutional nucleus. Experiments on the China traffic sign detection dataset (CCTSDB) show that the proposed algorithm enhances the ability to recognize small targets. Compared with the YOLOv7 model, the proposed algorithm has an average improvement of 5.22% in mAP and 9.01% in Recall, making it an effective traffic sign detection algorithm.

    • Task Diffusion of Mobile Crowdsensing in Social Network

      2023, 32(10):166-174. DOI: 10.15888/j.cnki.csa.009255

      Abstract (595) HTML (988) PDF 1.73 M (1410) Comment (0) Favorites

      Abstract:Mobile crowdsensing is one of the core basic technologies in the digital construction of smart cities, and it is a hot research topic in the field of mobile computing. In recent years, although there have been many representative research results on mobile crowdsensing, there is still a long way to go before it is widely used on a large scale, and it still faces the problem of low user participation in the actual promotion and application. To this end, the social network influence maximization (IM) transmission model is introduced. It considers the lack of probabilistic information in reality and learns the probability of influence while performing influence activities through online learning, or in other words, the influence model information is constantly updated according to the user feedback, so as to propose a new task diffusion scheme based on the model. Through experiments with real social network data sets, the results show that the proposed method is more effective than the traditional IM method in terms of transmission scope, and it makes a contribution to the practical promotion and application of mobile crowdsensing systems.

    • Efficient ZD Codes with Low Storage Based on ZigZag Decoding Algorithm Optimization

      2023, 32(10):175-183. DOI: 10.15888/j.cnki.csa.009275

      Abstract (522) HTML (801) PDF 1.57 M (1185) Comment (0) Favorites

      Abstract:ZigZag-decodable (ZD) codes, as a type of erasure codes, are designed and generated based on the ZigZag decoding algorithm. They only require a small computation overhead to repair failed data in the storage system but need to store more redundant data than other erasure codes to ensure the high reliability of the system. To reduce the storage overhead generated by the ZD codes, this study proposes an optimized ZigZag decoding algorithm by analyzing the idea of ZigZag decoding currently used in storage systems. The new decoding algorithm can make full use of the information in the parity data to repair data. This study also proposes a new ZD code encoding scheme based on the new decoding algorithm. Due to the higher information utilization of the new algorithm, the new encoding scheme can satisfy the high reliability of the storage system with less storage overhead. The experimental results show that the new ZD code encoding scheme proposed in this study has the optimal storage overhead, and the decoding and encoding performance is much higher than that of the widely used RS code.

    • Disease Risk Prediction Based on Label Noise Robust Learning

      2023, 32(10):184-191. DOI: 10.15888/j.cnki.csa.009268

      Abstract (740) HTML (810) PDF 1.60 M (1350) Comment (0) Favorites

      Abstract:Disease risk prediction enables the screening of vulnerable populations and early preventive interventions to reduce disease incidence and mortality. With the rapid development of machine learning technologies, disease risk prediction based on machine learning has been widely used. However, machine learning is highly dependent on high-quality labeling information, and the label noise in medical data will bring severe challenges to the construction of high-performance disease risk prediction algorithms. In order to solve this problem, a noise robustness learning method based on a deep neural network and dynamic truncation loss function is proposed for disease risk prediction. The dynamic truncation loss function is introduced in this method, which combines the implicit weighting characteristics of the traditional cross entropy function and the label noise robustness of the mean square error loss function. By constructing a training loss lower bound and introducing a dynamic sample weighting mechanism to reduce the gradient of suspicious samples, the weight of possible noisy samples in the training process is limited, and the robustness of the model is further enhanced. By taking the stroke screening dataset as an example, the experimental results show that the proposed algorithm can achieve excellent prediction performance under each ratio of label noises, reduce the negative impact of label noises in disease risk prediction, and realize robust learning of data with label noises.

    • Improved Marine Predator Algorithm and Its Application in WRSN Charging Planning

      2023, 32(10):192-200. DOI: 10.15888/j.cnki.csa.009246

      Abstract (547) HTML (789) PDF 3.11 M (1364) Comment (0) Favorites

      Abstract:For the multi-UAV charging planning under the wireless rechargeable sensor network, only considering the flight distance of the UAV to plan the optimal charging path for the cost target is single one-sided. Now the UAV flight distance, energy consumption, time cost, and UAV matching cost are combined into a new cost target model. To reduce the number of flight stops, a regular hexagonal charging model is also added, and an improved marine predator algorithm (BMPA) is proposed to be applied to this scenario. The improvement is as follows. On one hand, beetle antennae search algorithm is introduced into the marine predator algorithm to find the point with the largest odor value, which improves the optimal solution quality. On the other hand, a new adaptive nonlinear moving step parameter is added to the marine predator algorithm. As a result, the balance of exploration and development, and the global search ability are improved, and the rapid convergence of local research is promoted. The simulation results show that the proposed algorithm not only effectively reduces the number of flights, but also decreases the flight distance and computing power consumption. In addition, the new cost objective function values are reduced by 50.90%, 4.85%, and 14.38% compared with BAS, MPA, and PreWBAS algorithms, which proves the effectiveness of the improved algorithm.

    • Rain Removal Algorithm for Single Image Based on Expansive Corrosion Convolution Block

      2023, 32(10):201-207. DOI: 10.15888/j.cnki.csa.009223

      Abstract (537) HTML (743) PDF 1.88 M (1501) Comment (0) Favorites

      Abstract:When imaging equipment takes images on rainy days, the image quality will be seriously degraded due to the existence of rain fog and rain stripes, which will greatly affect the subsequent image processing performance. Therefore, the research on image rain removal algorithms has attracted wide attention, and the rain removal algorithm for a single image is facing great challenges because it is not supported by prior knowledge. In recent years, deep learning has been applied to the research on image rain removal algorithms because of its high feature representation ability. In this study, based on wavelet transform, an algorithm combining deep learning with morphological processing of digital images is adopted to remove rain from a single image, which has the advantages of a few training parameters, short training time, and good rain removal effects. Firstly, the image containing rain is decomposed into a low-frequency component, horizontal high-frequency component, vertical high-frequency component, and diagonal high-frequency component by wavelet transform. Then, the four components are constructed into deep learning neural networks respectively, and morphological processing such as image expansion and corrosion is added to the neural network architecture according to the rain features to remove rain, which greatly simplifies the model architecture and can achieve good results.

    • Color Constancy Calculation Based on Improved SqueezeNet

      2023, 32(10):208-214. DOI: 10.15888/j.cnki.csa.009232

      Abstract (600) HTML (1159) PDF 1.36 M (1497) Comment (0) Favorites

      Abstract:Due to the defects of imaging equipment, it is easy to cause the shift of the imaging color and affect the downstream tasks of the image algorithm. Therefore, the color constancy algorithm is required to correct the image color so that the image color is consistent with the color seen by the human naked eye. The effect of the traditional color constancy algorithm depends on specific light source environments. In order to improve the application range and utilization efficiency of the color constancy algorithm, a color constancy calculation model based on the SqueezeNet framework is proposed, which senses the image light source through the convolutional image network. In addition, the attention mechanism and residual connection are introduced to improve the network's understanding of images and computing performance. The network predicts the illumination color of each area of the input image at the same time and then gathers them by designing three different pooling methods to output the global estimated light source of the image, and it finally uses the estimated light source to correct the image. Experimental results show that the proposed light source estimation algorithm can effectively estimate the illumination color of the image and correct the image color.

    • Image Inpainting Algorithm Based on Generative Adversarial Network

      2023, 32(10):215-221. DOI: 10.15888/j.cnki.csa.009257

      Abstract (561) HTML (1097) PDF 2.45 M (1371) Comment (0) Favorites

      Abstract:The current depth learning network models based on generative adversarial networks encounter artifacts, texture detail degradation, and other phenomena when facing more complex features, which leads to a visual deficiency. In order to solve these problems, an improved image inpainting algorithm combining a generative adversarial network with a coherent semantic attention mechanism is proposed. First of all, the generator adopts a two-stage inpainting method, uses gated convolution to replace the ordinary convolution of the generative adversarial network, and introduces residual blocks to solve the gradient vanishing problem and a coherent semantic attention mechanism to enhance the generator's attention to the important information and structure in the image. Secondly, Markov discriminator is adopted to enhance the network's discrimination effect, and the output results of the generator are processed by deconvolution to get the final repaired image. By comparing the inpainting results and image quality evaluation indicators with the baseline algorithm, the experimental results show that the algorithm can better predict the missing parts and improve the inpainting effect.

    • Fast Waveform Design Algorithm for Precision Interference

      2023, 32(10):222-228. DOI: 10.15888/j.cnki.csa.009093

      Abstract (464) HTML (877) PDF 1.57 M (1329) Comment (0) Favorites

      Abstract:Precision interference is a new concept that receives widespread attention in the field of electronic warfare, which aims to avoid unexpected electromagnetic injuries to the ally's devices in the process of electronic interference. Its core idea is to realize the precise control of interference energy in the airspace. However, the current waveform design algorithm used for precision interference has the problem of high computational complexity, which can easily delay the time of electronic warfare. Given this problem, this study proposes a fast waveform design algorithm for precision interference. The optimization problem is established according to the precision interference spatial model. The upper bound of the targeted function is derived by adopting the Majorization-Minimization framework, and the available waveform for precision interference is obtained by iteratively solving the corresponding subproblems with the closed-form solutions. Simulation experiments demonstrate that the proposed algorithm has good performance on the indicators of precision interference and the superiority of lower computational complexity.

    • Fusion of Trigger Word Features for Event Extraction

      2023, 32(10):229-234. DOI: 10.15888/j.cnki.csa.009097

      Abstract (507) HTML (1251) PDF 970.25 K (1454) Comment (0) Favorites

      Abstract:Event extraction is a key research area in information extraction. To improve the effect of event extraction and solve the problem that general event extraction methods cannot make full use of text feature information, an event extraction method fused with trigger word features is proposed. A remote trigger word database is constructed to provide additional feature information for the event classification model and enhance the discovery ability of event trigger words. Then, the event type and the distance features of trigger words are integrated to improve the representation and learning ability of the event element extraction model. Finally, the event classification model and the event element extraction model are connected in series to improve the event extraction effect. Experiments on the DuEE dataset demonstrate that compared with other models, this model improves the accuracy, recall, and F1 value, which proves the effectiveness of this model.

    • Distilling Inter-class Distance for Semantic Segmentation

      2023, 32(10):235-241. DOI: 10.15888/j.cnki.csa.009276

      Abstract (599) HTML (867) PDF 1.63 M (1202) Comment (0) Favorites

      Abstract:Knowledge distillation is widely adopted in semantic segmentation to reduce the computation cost. The previous knowledge distillation methods for semantic segmentation focus on pixel-wise feature alignment and intra-class feature variation distillation, neglecting to transfer the knowledge of the inter-class distance, which is important for semantic segmentation. To address this issue, this study proposes an inter-class distance distillation (IDD) method to transfer the inter-class distance in the feature space from the teacher network to the student network. Furthermore, since semantic segmentation is a position-dependent task, thus this study exploits a position information distillation module to help the student network encode more position information. Extensive experiments on three popular semantic segmentation datasets: Cityscapes, Pascal VOC, and ADE20K show that the proposed method is helpful to improve the accuracy of semantic segmentation models and achieves great performance.

    • Node Importance Evaluation of Combat System Based on Operation Loop

      2023, 32(10):242-254. DOI: 10.15888/j.cnki.csa.009094

      Abstract (603) HTML (977) PDF 2.31 M (2074) Comment (0) Favorites

      Abstract:Taking the actual combat system as the research object, this work studies the modeling and evaluation methods of combat systems theoretically based on the complex network theory and modern combat cycle theory. Firstly, an investigation, decision, fire, communication, and support (IDFCS) combat model based on functional combat model is proposed, with its modeling method and generation algorithm introduced, which can quantitatively describe system armament capability and combat system rules from a more realistic and detailed perspective. Secondly, this work introduces a system effectiveness evaluation method and armament importance evaluation method based on IDFCS model and operation loop theory, which can quantitatively analyze the effectiveness of combat system by using the capability attributes of armaments. Finally, the work generates the combat system sample networks of different scales with the model's generation algorithm and evaluates its effectiveness, which provides theoretical support for further research of system operations.

    • Object-centric Command and Control Process Modeling and Its Application

      2023, 32(10):255-264. DOI: 10.15888/j.cnki.csa.009096

      Abstract (589) HTML (984) PDF 1.92 M (1519) Comment (0) Favorites

      Abstract:Process modeling is the key step to analyzing the command and control process and winning the battle. The current modeling methods cannot well deal with the characteristics of multi-domain association and loose-coupled correlation of command and control processes in the joint battle. In view of the above problems, this study proposes an object-oriented process modeling language. The built model describes the data perspective, behavior perspective, and the complex interaction between them in the command and control process. Based on this model, process mining technology is applied to restore the command and control process, which provides an effective means to optimize the command and control process and improve process efficiency.

    • Community Detection and Link Prediction Based on Network Embedding Stochastic Blockmodel

      2023, 32(10):265-274. DOI: 10.15888/j.cnki.csa.009254

      Abstract (576) HTML (1143) PDF 1.78 M (1469) Comment (0) Favorites

      Abstract:Community detection and link prediction are hot issues in network data research. Taking into account both network transitivity and block structure can help capture the effective association between individuals and detect the inherent patterns in the data, thus helping researchers explore more data values and make decisions. Most of the current algorithms and models focus on single-level analysis of network transitivity or block structure, and they rely on certain assumptions This study proposes a network embedding stochastic blockmodel (NE-SBM) for community detection and link prediction. A Bayesian framework is built to regularize the model parameters, and the Metropolis Hasting-Gibbs algorithm is applied to obtain the hidden location and community affiliation represented by node embedding. The study also takes advantage of the multidimensional scaling algorithm to solve the hidden location identifiability problem. The proposed method can solve the problem of over-reliance on judgment criterion or evaluation function in traditional heuristic algorithms and has better adaptability to all types of data. In addition, the experimental results on artificial and real data further validate the superior performance of the method in community detection and link prediction.

    • Medical Image Classification Based on Predictive Adversarial Networks

      2023, 32(10):275-283. DOI: 10.15888/j.cnki.csa.009270

      Abstract (618) HTML (748) PDF 1.53 M (1182) Comment (0) Favorites

      Abstract:Positive-unlabeled learning (PU learning) only uses unlabeled samples and positive samples to train a binary classifier, while generative adversarial networks (GANs) obtain an image generator through adversarial training. In order to transfer the adversarial training method of GANs to PU learning for higher PU learning performance, the generator in GANs can be replaced with a classifier C, which selects samples in the unlabeled dataset to deceive the discriminator D and optimize C and D iteratively. This study proposes the JS-PAN model, which uses the Jensen-Shannon divergence (JS-divergence) as the objective function. Finally, according to the characteristics of data distribution and current needs, the rationality and high performance of the PAN model applied in the binary classification of medical diagnostic images are explained. Experiments on MNIST and CIFAR-10 datasets show that the KL-PAN model has higher accuracy (ACC) and F1-score than the similar PU learning models, and the JS-PAN model has higher performance in terms of two indicators after symmetric improvement, so the JS-PAN model is more reasonable. Experiments on three image subdatasets of Med-MNIST show that the KL-PAN model has almost the same ACC as the four benchmark supervised models, and JS-PAN has higher performance. Therefore, in view of both the excellent classification performance of the PAN model and the distribution characteristics of medical diagnostic data, PAN, as a semi-supervised learning method, can achieve faster and better results and thus show higher performance in the task of binary classification of medical images.

    • Emotion Recognition of EEG Signals Based on SAE and GNDO-SVM

      2023, 32(10):284-292. DOI: 10.15888/j.cnki.csa.009264

      Abstract (553) HTML (916) PDF 3.87 M (1304) Comment (0) Favorites

      Abstract:Affective computing is a key problem in modern human-computer interaction, and with the development of artificial intelligence, emotion recognition based on electroencephalogram (EEG) has become an important research direction. To improve the classification accuracy of emotion recognition, this study introduces stacked auto-encoder (SAE) to extract the deep feature of EEG multichannel signals and then proposes a generalized normal distribution optimization based support vector machine (GNDO-SVM). The experimental results show that the proposed GNDO-SVM model has better classification performance than the support vector machine model optimized by genetic algorithm, particle swarm optimization algorithm, and sparrow search algorithm. The accuracy of emotion recognition based on SAE depth features reaches 90.94%, indicating that SAE can effectively exploit the depth correlation information between different channels of EEG signals. Therefore, applying SAE depth feature extraction combined with the GNDO-SVM classification model can effectively achieve the emotion recognition of EEG signals.

    • Grammatical Error Correction Model Based on Differential Fusion Syntactic Feature

      2023, 32(10):293-300. DOI: 10.15888/j.cnki.csa.009259

      Abstract (518) HTML (925) PDF 1.36 M (1083) Comment (0) Favorites

      Abstract:Current English GEC methods tend to ignore the syntactic knowledge in texts, which plays an important role in grammatical error correction, and thus the error correction ability of English GEC models is affected. To address this problem, the study proposes a GEC method which is based on the differential fusion syntactic features. First, the proposed syntactic encoder can generate dependency graph and constituency syntactic tree information from raw data in an unsupervised way and conduct the feature fusion of these two heterogeneous syntactic structures to encode high-dimensional syntactic representation. Second, to utilize both semantic and syntactic information in the text, the differential fusion module first uses differential regularization to enhance the semantic encoder to capture the semantic features that the syntactic encoder fails to generate. Then the syntactic representation and semantic representation are further fused by cross attention as the output features of the Transformer encoder, which are finally input to the decoder to generate grammatically correct text. The comparison experiment on the CoNLL-2014 task dataset shows that the precision and F0.5 value of this method are better than those of the GEC model based on the Copy-Augmented Transformer, and the F0.5 value of this method is improved by 5.2 percentage points. The syntactic knowledge avoids the problem of lacking high-quality annotated training corpora and has a better performance in text error correction.

    • Identity-based Proxy Signature on Lattices

      2023, 32(10):301-307. DOI: 10.15888/j.cnki.csa.009243

      Abstract (654) HTML (946) PDF 1.22 M (1332) Comment (0) Favorites

      Abstract:To resist quantum computing attacks and reduce the risk of private key leakage of users in proxy signatures, this study proposes an identity-based proxy signature scheme on lattices. This scheme is designed based on the secure and efficient GPV signature framework. The verification public key is generated by combining the user identity information. The lattice basis delegation technology is used to generate the private key for the user signature, and the bonsai tree delegation algorithm is adopted to improve signing efficiency. The security of the scheme is based on the shortest integer solution (SIS) assumption. It satisfies the security properties of identity-based proxy signatures and has existential unforgeability under random oracles and quantum random oracles.

Current Issue


Volume , No.

Table of Contents

Archive

Volume

Issue

联系方式
  • 《计算机系统应用》
  • 1992年创刊
  • 主办单位:中国科学院软件研究所
  • 邮编:100190
  • 电话:010-62661041
  • 电子邮箱:csa (a) iscas.ac.cn
  • 网址:http://www.c-s-a.org.cn
  • 刊号:ISSN 1003-3254
  • CN 11-2854/TP
  • 国内定价:50元
You are the firstVisitors
Copyright: Institute of Software, Chinese Academy of Sciences Beijing ICP No. 05046678-3
Address:4# South Fourth Street, Zhongguancun,Haidian, Beijing,Postal Code:100190
Phone:010-62661041 Fax: Email:csa (a) iscas.ac.cn
Technical Support:Beijing Qinyun Technology Development Co., Ltd.

Beijing Public Network Security No. 11040202500063