Most Download

Published in last 1 year | In last 2 years| In last 3 years| All| Most Downloaded in Recent Month | Most Downloaded in Recent Year|

In last 3 years

Please wait a minute...


For Selected: Download Citations EndNote Ris BibTeX Toggle Thumbnails

Select

Review of Neural Network Pruning Techniques JIANG Xiaoyong, LI Zhongyi, HUANG Langyue, PENG Mengle, XU Shuyang Journal of Applied Sciences    2022, 40 (5): 838-849.   DOI: 10.3969/j.issn.0255-8297.2022.05.013 Abstract （3278）      PDF（pc） （1569KB）（1920）       Knowledge map    Save This paper summaries the origin and research progress of neural network pruning technologies, divides them into two categories of unstructured pruning with sparse weight parameters and coarse-grained structured pruning, and introduces the representative methods of the two categories in recent years. Because pruning reduces model parameters and compresses the model size, depth models can be applied to embedded devices, showing the importance of pruning in the field of deep learning model compression. In view of the existing pruning technologies, this paper expounds the problems existing in practical applications and measurement standards, and prospects the research and development tendency in the future. Reference | Related Articles | Metrics | Comments（0）

Select

S-Type Fiber Cladding SPR Sensor for Hg2+ Concentration Detection WEI Yong, ZHAO Xiaoling, WANG Rui, LIU Chunlan, ZHANG Yonghui Journal of Applied Sciences    2022, 40 (6): 896-905.   DOI: 10.3969/j.issn.0255-8297.2022.06.002 Abstract （1974）      PDF（pc） （6185KB）（1204）       Knowledge map    Save Fiber cladding surface plasmon resonance (SPR) sensors can be employed in biochemical sensing without the requirement of cladding removal, but with the limitation of lack of architectural variety. In this paper, a new type of cladding SPR sensor construction based on S-type fiber is proposed. The S-type fiber is first fabricated by electrofusion technology, and then coated with a 50 nm gold film on its cladding surface. In the S-type fiber, light is coupled from the fiber core to the cladding, and contacted with the gold film on the cladding surface to form a fiber cladding SPR sensor, thus, realizing the detection of Hg2+ molar concentration. The resonance valley wavelength of the SPR sensor changes from 633.03 nm to 645.09 nm as the Hg2+ molar concentration increasing from 5 pmol/L to 5 mmol/L. The detection sensitivity reaches 1.382 nm/lg(mmol/L). The manufacture of S-type fiber cladding SPR sensor features with simplicity and can be used for single-mode and multi-mode optical fibers to realize cladding SPR sensors which are especially useful in the field of biochemical detection. Reference | Related Articles | Metrics | Comments（0）

Select

Image Privacy Protection Based on Cycle-Consistent Generative Adversarial Networks XIE Yiyi, ZHANG Yushu, ZHAO Ruoyu, WEN Wenying, ZHOU Yuqian Journal of Applied Sciences    2023, 41 (2): 228-239.   DOI: 10.3969/j.issn.0255-8297.2023.02.004 Abstract （2590）      PDF（pc） （9135KB）（821）       Knowledge map    Save Social media and cloud computing have facilitated the distribution and storage of images. Meanwhile, concerns about image privacy have been raised. It is crucial to protect image privacy from privacy violation and illegal use. Motivated by this, an image privacy protection method based on cycle-consistent generative adversarial networks (CycleGAN) is proposed in this paper. Considering the usability in image privacy protection, the method first combines image segmentation with CycleGAN to select different segmentation coefficients to generate images with different degrees of privacy protection. Then reversible information hiding is used to embed information in the generated privacy preserving image, so as to prevent illegal users from reversing the image. Thus, a balance is achieved between image privacy protection and usability. The proposed method is trained and tested using PIPA dataset, using peak signal to noise ratio and structural similarity index are used as performance metrics to evaluate the privacy-preserving images. Experimental results show that the proposed method outperforms other schemes in both image privacy preservation and usability. Reference | Related Articles | Metrics | Comments（0）

Select

Fatigue Failure Model of IGBT Chip Based on Threshold Voltage LI You, CAO Jiwei, HAO Guangyao, YAN Ge, LIU Hongxiao Journal of Applied Sciences    2022, 40 (5): 865-875.   DOI: 10.3969/j.issn.0255-8297.2022.05.015 Abstract （1807）      PDF（pc） （1175KB）（760）       Knowledge map    Save In order to effectively evaluate the health status of IGBT during its whole life cycle, the fatigue failure mechanism of IGBT chip was studied based on the theory of semiconductor physics, and the effect of charge density at gate interface on threshold voltage was analyzed. Taking the threshold voltage as the failure characteristic quantity of IGBT, the fatigue failure model of IGBT chip was established on the basis of studying the change rule of threshold voltage with fatigue failure time. An IGBT threshold voltage test platform was built, and IGBT aging experiments were performed to verify that the model proposed in this paper can accurately characterize and estimate the aging degree of IGBT chips, and the correctness and rationality of the failure model were verified. Reference | Related Articles | Metrics | Comments（0）

Select

SlightDetection: A Static Analysis Tool for Smart Contracts Security Vulnerabilities on Ethereum CHEN Xiaohan, ZHAO Xiangfu, ZHANG Dengji, FEI Jiajia Journal of Applied Sciences    2022, 40 (4): 695-712.   DOI: 10.3969/j.issn.0255-8297.2022.04.012 Abstract （1966）      PDF（pc） （645KB）（745）       Knowledge map    Save Security vulnerabilities in Ethereum smart contracts may lead to immeasurable losses. To alleviate this problem, a smart contract vulnerability detection tool SlightDetection is proposed, which uses static program analysis technology to achieve full code coverage. The tool converts smart contract source codes into a corresponding abstract syntax tree, and translates it into an XML intermediate representation. Taking the characteristics of several classic vulnerabilities as an example, the tool writes a custom XPath rule library, and using the XML intermediate representation and the XPath library as inputs, the tool keeps traversing and matching the XPath rule base, till getting the report of vulnerability detection. This work tests 3 classic contracts, and fully demonstrates the faster and more accurate detection features of SlightDetection. The effectiveness of the tool is proved by testing a large number of smart contracts provided on Etherscan and manually verifying more than 100 of them. Reference | Related Articles | Metrics | Comments（0）

Select

Stock Price Trend Prediction Based on Dual-Stream LSTM Neural Network WU Feng, XIE Cong, JI Shaopei Journal of Applied Sciences    2023, 41 (2): 344-358.   DOI: 10.3969/j.issn.0255-8297.2023.02.013 Abstract （2279）      PDF（pc） （1900KB）（639）       Knowledge map    Save Previous research on stock price volatility prediction relies on analyzing shallow features of financial news datasets and ignores the structural relationship between words in financial news, resulting in poor prediction performance. Aiming at this problem, we propose a stock price trend prediction model (Sent2Vec-DLSTM) based on a dual-stream long short-term memory network (LSTM) neural network. A vector generation model of emotional words called Sent2Vec is first proposed based on financial stock news data set and Harvard IV-4 emotion dictionary training, which is then combined with dual-stream LSTM neural network (DLSTM). In the experiment, the historical data of the S&P 500 index and the financial articles obtained by crawling are used to predict the trend of the S&P 500 index. the VietStock news and stock price data from cophieu68 are then used to predict the trend of the VN index. The results show that Sent2Vec-DLSTM outperforms existing models in stock price trend prediction. Reference | Related Articles | Metrics | Comments（0）

Select

Multi-modal Emotion Recognition Using Speech, Text and Motion JIA Ning, ZHENG Chunjun Journal of Applied Sciences    2023, 41 (1): 55-70.   DOI: 10.3969/j.issn.0255-8297.2023.01.005 Abstract （2710）      PDF（pc） （1376KB）（577）       Knowledge map    Save For the problems of low accuracy and weak generalization ability in the process of human emotion recognition, a fusion method of multi-modal emotion recognition based on speech, text and motion is proposed. In the speech mode, a depth wavefield extrapolation-improved wave physics model (DWE-WPM) is designed to simulate the sequence information mining process of long short-term memory (LSTM) network; In the text mode, a transformer model with multi-attention mechanism is used to capture the potential semantic expression of emotion; In the motion mode, sequential features of facial expression and hand action are combined by using two-way three-layer LSTM model with attention mechanism. Accordingly, a multi-modal fusion scheme is designed to achieve high-precision and strong generalization ability of emotion recognition. In the general emotion corpus IEMOCAP, the method proposed in this paper is compared with existing emotion recognition algorithms. Experimental results show that the proposed method has higher recognition accuracy both in single modality and multi-modals, with average accuracy improved by 16.4% and 10.5% respectively, effectively improving the ability of human emotion recognition in human-computer interaction. Reference | Related Articles | Metrics | Comments（0）

Select

Architecture Design and Implementation of Seafood-Oriented Supply Chain Platform Based on Multiple Heterogeneous Blockchains WU Ou, ZHANG He, WANG Yanze, LI Haoming, LI Shanshan Journal of Applied Sciences    2022, 40 (4): 539-554.   DOI: 10.3969/j.issn.0255-8297.2022.04.001 Abstract （2224）      PDF（pc） （1509KB）（536）       Knowledge map    Save Blockchain technology can be used to solve the problems of data fraud, low transparency and difficulty in tracking in traditional seafood supply chains. However, if organizations in a supply chain adopt a heterogeneous blockchain which is unable to exchange data and complement each other's functions, there will be value islands. To solve this problem, this paper chooses BitXHub cross-chain solution based on relay mechanism, then designs and implements a cross-border seafood supply chain platform that supports the interoperation of mainstream blockchain Hyperledger Fabric and Ethereum. Experimental results show that the system can process more than 937 transactions per second, which can meet the business requirements of actual scenes. Reference | Related Articles | Metrics | Comments（0）

Select

Research Progress on Frequency Diversity Array Radar: from System Framework to Parameter Estimation ZHANG Xiaofei, WANG Cheng, LI Jianfeng, WU Qihui Journal of Applied Sciences    2022, 40 (6): 918-940.   DOI: 10.3969/j.issn.0255-8297.2022.06.004 Abstract （1858）      PDF（pc） （849KB）（522）       Knowledge map    Save The received signal of frequency diversity array (FDA) is a function of the angle and range of target by using little frequency offsets among array units in transmitter, which can be utilized to conduct beamforming, joint angle and range estimation and interference suppression. In this paper, the research background and progress are presented, and the principle and signal model of parameter estimation utilizing FDA-MIMO radar are introduced. After that, three frequency offset frameworks, including unfolded coprime linear array-unfolded coprime frequency offset (UCLA-UCFO), virtual coprime planar arrayunfolded coprime frequency offset (VCPA-UCFO), and synthetic aperture multi-coprime frequency offset (SA-MCFO) frameworks are proposed. In addition, several new works on algorithms, including reduce dimension root multi-signal classification (RD-root-MUSIC), fast convergence-trilinear decomposition (FC-TD), successive iteration (SUIT), twice reduce dimension-MUSIC (TRD-MUSIC) and reduce dimension MUSIC with decoupling (RDMD) algorithms are also presented. Finally, the effectiveness and superiority of the frameworks and algorithms are verified by simulations. Reference | Related Articles | Metrics | Comments（0）

Select

Fault Analysis and Retrieval of Message Based on Knowledge Graph JI Wenlu, DENG Xing, ZHU Hongqin, ZHAO Yang, JIANG Jielin Journal of Applied Sciences    2023, 41 (3): 378-390.   DOI: 10.3969/j.issn.0255-8297.2023.03.002 Abstract （973）      PDF（pc） （1490KB）（511）       Knowledge map    Save Aiming at the problems of complex fault analysis mode and increasing difficulty of fault removal caused by the expansion of the power grid, a fault analysis and retrieval of message based on knowledge graph is proposed. Firstly, the construction of power dispatching fault knowledge graph is completed by combining natural language processing technology with expert documents. Then the expert knowledge is stored in the graph in the form of atomic rules (indivisible rules) to achieve intelligent fault analysis and retrieval and assist maintenance personnel in decision-making, thereby improving the efficiency of business process. Finally, combined with the knowledge graph and log information, artificial intelligence is used to analyze the cause of failure, and the optimal solution is obtained from multiple potential solutions. Experimental results on both real and synthetic data sets show that the proposed method can achieve good results on fault analysis and retrieval in power dispatching. Reference | Related Articles | Metrics | Comments（0）

Select

Review of Steganalysis for Digital Images WANG Zichi, LI Bin, FENG Guorui, ZHANG Xinpeng Journal of Applied Sciences    2024, 42 (5): 723-732.   DOI: 10.3969/j.issn.0255-8297.2024.05.001 Abstract （538）      PDF（pc） （2695KB）（503）       Knowledge map    Save Digital steganography plays a crucial role in securely transmitting confidential data by concealing it within common multimedia, such as images, videos, and audio, to facilitate covert communication. To discover the covert communication of steganography, the technique of steganalysis can be employed. Steganalysis determines whether a given multimedia object contains secret data according to the statistical anomaly of stego data caused by steganography. In recent years, both steganography and steganalysis have made significant progress and development in their mutual confrontation, particularly in the context of the growing prevalence of digital images on social networks. Focusing on digital images, this paper sorts out the development of steganalysis in the past decade, and reviews the traditional steganalysis and deep learning based-steganalysis. Then, the limitations of each approach are discussed. Finally, the study offers insights into the prospective development trends in steganalysis. Reference | Related Articles | Metrics | Comments（0）

Select

Traceable DNN Model Protection Based on Watermark Neural Network LIU Yalei, HE Hongjie, CHEN Fan, LIU Zhuohua Journal of Applied Sciences    2023, 41 (2): 183-196.   DOI: 10.3969/j.issn.0255-8297.2023.02.001 Abstract （2489）      PDF（pc） （9105KB）（501）       Knowledge map    Save This paper proposes a multi-user traceability watermarking neural network approach to model security and copyright certification for deep neural networks (DNN). The watermark is generated by the key driver and embedded invisibly in the output images of the DNN model, hence realizing the intellectual property protection and copyright tracking of DNN model. A codec network is added to the DNN model to embed the watermark, and a two-stream tamper detection network is used as the discriminator. Thus, the problem of residual watermark in the output images of the model is solved, which, reduces the impact on the performance of DNN model and enhances the security. In addition, a two-stage training method is proposed in this paper to distribute different watermarked models to different users. When copyright disputes occur, another residual network can be used to extract the watermark image from the output image. Experiments show that the proposed method is efficient in distributing watermarked models, and is able to trace the source of DNN models embedded with similar watermarked images for multiple users. Reference | Related Articles | Metrics | Comments（0）

Select

Evaluation Method of Blockchain Privacy Protection for Full Life Cycle of Transaction Data ZHU Xuguang, XING Chunxiao, LI Wenqing, HAO Yingting Journal of Applied Sciences    2022, 40 (4): 555-566.   DOI: 10.3969/j.issn.0255-8297.2022.04.002 Abstract （1981）      PDF（pc） （768KB）（493）       Knowledge map    Save This paper proposes an evaluation method of blockchain privacy protection for full life cycle of transaction data in view of the differences between blockchain and traditional information systems, analyzes blockchain privacy leakage risk and privacy protection methods from aspects of transaction data that are release, consensus, storage, and application to establish a blockchain privacy protection evaluation index system; conduct relative importance decision-making for index, and weight calculation using a dynamic index weight assignment method combining analytic hierarchy process and pairwise comparison; combine three dimensions of privacy protection strength, transaction data usability and privacy protection technical performance to calculate blockchain privacy protection related capability score. The analysis shows that the method enables a comprehensive evaluation of the level of blockchain privacy protection. Reference | Related Articles | Metrics | Comments（0）

Select

Multi-modal Diagnosis Method of Alzheimer’s Disease LI Weihan, HOU Beiping, HU Feiyang, ZHU Bihong Journal of Applied Sciences    2023, 41 (6): 1004-1018.   DOI: 10.3969/j.issn.0255-8297.2023.06.008 Abstract （465）      PDF（pc） （3577KB）（492）       Knowledge map    Save The current grading methods for Alzheimer’s disease（AD）, Early Mild Cognitive Impairment（EMCI）, and Normal Control（NC） suffer from difficulties recognizing EMCI and low multi-classification accuracy. To address these issues, a brain region feature extraction method is proposed, and an AD multi-modal classification model is designed with a fusion of ResNet network. Brain MRI images are spatially registered, segmented by Bayesian and Gaussian mixture models to obtain gray matter, the regions with the greatest difference are selected as the feature image area, and images and biomarkers are processed by the classification model. The proposed method improves performance by at least 5% and achieves an accuracy of 95.5%, 93.5%, and 86.3% for AD＆NC, AD＆EMCI,and AD＆EMCI＆NC classification, respectively, surpassing any single-modal network and verifying the effectiveness of this method. Reference | Related Articles | Metrics | Comments（0）

Select

Building Change Detection in Remote Sensing Images Based on Semantic Segmentation YIN Meijie, NI Cui, WANG Peng, ZHANG Guangyuan Journal of Applied Sciences    2023, 41 (3): 448-460.   DOI: 10.3969/j.issn.0255-8297.2023.03.007 Abstract （1168）      PDF（pc） （6613KB）（488）       Knowledge map    Save Remote sensing image change detection is to use multi-temporal images to determine the changes of objects or phenomena within a certain period of time, and to provide qualitative and quantitative information on spatial changes of objects. Traditional remote sensing image change detection methods are mainly based on ground texture and spatial features, which is difficult to accurately identify new buildings in remote sensing images. Therefore, this paper adopts a building change detection method based on UNet network. Firstly, the lightweight efficient channel attention network (ECANet) is injected into the original UNet network model to adjust and optimize the network structure and improve the accuracy of image segmentation. The parameters of SENet are then tuned to enhance the accuracy of building change detection in remote sensing images. Experiments on a high-resolution dataset LIVER-CD show that the accuracy of the proposed method reaches a semantic segmentation accuracy of 99.03% and a building change detection accuracy of 98.62%. Compared with other methods, the proposed method can effectively enhance the effective features of images and improve the detection accuracy of ground buildings in remote sensing images. Reference | Related Articles | Metrics | Comments（0）

Select

Land Cover Classification of Sentinel-2 Image Based on Multi-feature Convolution Neural Network HUANG Xianpei, MENG Qingxiang Journal of Applied Sciences    2023, 41 (5): 766-776.   DOI: 10.3969/j.issn.0255-8297.2023.05.004 Abstract （450）      PDF（pc） （1320KB）（486）       Knowledge map    Save The 10 m resolution Sentinel-2 images takes the spectral values of the image as input in the original GoogLeNet without treating the ground objects in the image as a whole. To leverage object-oriented features in Sentinel-2 remote sensing image classification, this paper proposes an Object-oriented GoogLeNet network structure based on multiple features. Object-oriented GoogLeNet incorporates object-oriented spectral and shape features, and fully utilizes the shape features of differences between different ground objects for classification. On the data set of cloudless images in Wuhan and its surrounding areas, the overall accuracy of the classification results of Object-oriented GoogLeNet model has increased by 1.773% compared to GoogLeNet. The results show that the model with object-oriented features enhances the classification performance of Sentinel-2 remote sensing images. Reference | Related Articles | Metrics | Comments（0）

Select

Cross-Modal Person Re-identification Driven by Cross-Channel Interactive Attention Mechanism in Dual-Stream Networks HE Lei, LI Fengyong, QIN Chuan Journal of Applied Sciences    2024, 42 (5): 884-892.   DOI: 10.3969/j.issn.0255-8297.2024.05.014 Abstract （250）      PDF（pc） （4390KB）（485）       Knowledge map    Save Existing cross-modal person re-identification methods often fail to take into account the difference of target person between modes and within modes, making it difficult to further improve the retrieval accuracy. To solve this problem, this paper introduces the cross-channel interaction attention mechanism to enhance the robust extraction of person features, effectively suppresses the extraction of irrelevant features and achieves more discriminative feature expression. Furthermore, hetero-center triplet loss, triplet loss and identity loss are combined for supervised learning, effectively integrating the intermodal and intra-class differences in person features. Experimental results demonstrate the effectiveness of the proposed method, which outperforms seven existing methods on two standard datasets, RegDB and SYSU-MM01. Reference | Related Articles | Metrics | Comments（0）

Select

Experimental Evaluation of FSM Conformance Testing Based on Structure Coverage and State Identification LIN Weiwei, ZENG Hongwei, MIAO Huaikou, WANG Xiaolin Journal of Applied Sciences    2023, 41 (2): 326-343.   DOI: 10.3969/j.issn.0255-8297.2023.02.012 Abstract （2097）      PDF（pc） （786KB）（479）       Knowledge map    Save In finite state machine (FSM) conformance testing, there are two widely used test generation techniques which are based on structure coverage and state identification respectively. Under the condition of scarce test resources, we often face the problem of weighing selection of different test methods. To the best of our knowledge, there is no comprehensive comparative study of these two test techniques so far. This paper presents the necessity of experimental evaluation of the two test methods, and conducts experiments based on 10 FSM empirical cases. The performance is evaluated in terms of test cost and fault coverage capability, so as to provide empirical suggestions for the selection and application of these two techniques in FSM conformance testing. Reference | Related Articles | Metrics | Comments（0）

Select

Named Entity Recognition Algorithm Enhanced with Entity Category Information LIU Minghui, TANG Wangjing, XU Bin, TONG Meihan, WANG Liming, ZHONG Qi, XU Jianjun Journal of Applied Sciences    2023, 41 (1): 1-9.   DOI: 10.3969/j.issn.0255-8297.2023.01.001 Abstract （1803）      PDF（pc） （1423KB）（476）       Knowledge map    Save To solve the problem that the character level model of Chinese named entity recognition (NER) may ignore word information in sentences, a Chinese NER method based on entity category information enhancement in knowledge graph was proposed. Firstly, a training set was segmented with word segmentation tool, and all possible words were selected to construct a vocabulary. Secondly, the category information of entities in the vocabulary was retrieved by using generic knowledge graph, to construct a word set related to characters in a simple and effective way, and an entity category information set is generated according to the category information of entities in the word set. Finally, word embedding method was used to convert the set of category information into embeddings and concatenation of character embeddings, so as to enrich features in embedding layer. The proposed method can either be used as a module to expand feature diversity of embedding layer, or jointly applies with a variety of encoder-decoder models. Experiments on the Chinese NER dataset proposed by Microsoft Research Asia (MSRA) show the superiority of the proposed model. Compared with the models of Bi-directional long short-term memory (Bi-LSTM) and Bi-LSTM plus with conditional random field (CRF), the proposed method increases F1 by 11.00% and 3.09% respectively, verifying that the category information of entities in knowledge graph performs high effectiveness in the enhancement of Chinese NER. Reference | Related Articles | Metrics | Comments（0）

Select

Identification Method for Vessel Interrupt Track Correlating Based on Fuzzy Membership Degree CHEN Zhaotong, CHEN Jiangping, PAN Li Journal of Applied Sciences    2023, 41 (2): 296-310.   DOI: 10.3969/j.issn.0255-8297.2023.02.010 Abstract （2425）      PDF（pc） （2053KB）（473）       Knowledge map    Save In order to integrate the vessels’ tracks from different monitoring sources and form a unified maritime posture, a track correlation identification method based on track prediction and fuzzy membership evaluation is proposed. A polynomial fitting method is used to speculate the past and future tracks. Position, course, and speed are selected as the fuzzy factors. The membership degree of each fuzzy factor between the predicted track is calculated by ridge fuzzy membership function and weighted to obtain the membership of a single moment. A weighting function is constructed to calculate the comprehensive membership degree and finally the threshold is set to determine whether the two tracks are correlated. In simulation experiments, the precision and the recall rate of the proposed method is higher than 90% and 80% respectively, which outperforms the traditional method. In order to further investigate the applicability of the proposed method, the radar monitoring data of vessel tracks under different scenarios are simulated, including stable scenarios, speed change and course change scenarios during the interruption interval. Simulation results show that the proposed method provides an effective way to solve the problem of correlation identification of interrupted tracks in the cross environment of non-cooperative vessels. Reference | Related Articles | Metrics | Comments（0）