Table of Content

15 December 2022 Volume 35 Issue 12

  

An Online-Method of Multiple Change Points Detection Based on Random and Overlapping Strategy

ZHU Junjun,QI Jinpeng,ZHONG Jinmei,REN Qing,CAO Yitong

Electronic Science and Technology. 2022, 35(12):  1-9.  doi:10.16180/j.cnki.issn1007-7820.2022.12.001

Abstract ( 256 )   HTML ( 7 )   PDF (5527KB) ( 52 )  

Figures and Tables | References | Related Articles | Metrics

The traditional detection methods of multiple change points are mainly off-line, and cannot detect large-scale time series data online. To solve this problem, this study proposes an online detection method of multiple change points based on the buffer model and the sliding window random overlapping strategy. This method is based on TSTKS algorithm and sliding window model, receives online time series data stream in real time through buffer model, and transfers the data to the data receiver. Subsequently, the data stream is segmented using a sliding window random overlap strategy in the data sink. Finally, in the sub-data stream, TSTKS algorithm is used to perform online detection of multiple change points on the data. The experimental results of simulation data and EMG data of epilepsy patients show that the proposed method has the advantages of shorter time consumption and higher accuracy, and can be considered as an alternative for online analysis of large-scale time series data streams.

Classification and Recognition of P300 Event-Related Potential Based on LSTM-Attention Network

WANG Xialin,KAN Xiu,FAN Yixuan

Electronic Science and Technology. 2022, 35(12):  10-16.  doi:10.16180/j.cnki.issn1007-7820.2022.12.002

Abstract ( 356 )   HTML ( 8 )   PDF (1200KB) ( 60 )  

Figures and Tables | References | Related Articles | Metrics

In view of the problem of low recognition and classification accuracy of P300 event-related potentials in EEG signals, a recognition and classification method of P300 event-related potentials based on LSTM-Attention network is proposed in this study. In the data processing stage, SMOTE is utilized to augment P300 potential data in EEG signals, and irrelevant noise in synthetic data is eliminated based on DBSCAN clustering algorithm. In the identification and classification stage, an LSTM-Attention classification and identification network is built by adding an attention mechanism and a Dropout layer after the LSTM network, and the Sigmoid function is used to output the identification and classification results of the P300 event-related potential. The experimental results show that the proposed method can effectively recognize and classify P300 event-related potentials in EEG signals, and the average accuracy and Dice coefficient are up to 91.9% and 91.7%, respectively. Compared with traditional methods, the accuracy is higher and the generalization performance is stronger.

Knowledge Graph Query Method Based on Geographic Location Information

LI Yipei,WANG Yuxiang

Electronic Science and Technology. 2022, 35(12):  17-25.  doi:10.16180/j.cnki.issn1007-7820.2022.12.003

Abstract ( 379 )   HTML ( 5 )   PDF (2357KB) ( 37 )  

Figures and Tables | References | Related Articles | Metrics

Existing knowledge graph query methods ignore the geographic location information of entities themselves, so they do not support geographic location related queries. In view of this problem, on the basis of the hybrid knowledge graph integrating geographic location information, this study proposes a knowledge graph query method based on geographic location information. By extracting the triples from the query problem, the corresponding query graph is constructed to understand the natural language query problem. The query problems based on geographic location information are divided into six categories, and combined with the existing semantic query methods for fact-based problems, the corresponding knowledge graph query methods are studied according to the query graph or K-nearest neighbor search idea. Experimental results show that the accuracy rate of the proposed method can reach more than 77%, which can provide effective support for query based on geographic location information.

Surrogate Assisted Multi-Objective Particle Swarm Optimization Based on Combined Infill Sampling Criterion

CHEN Wanfen,WANG Yujia,LIN Weixing

Electronic Science and Technology. 2022, 35(12):  26-34.  doi:10.16180/j.cnki.issn1007-7820.2022.12.004

Abstract ( 314 )   HTML ( 6 )   PDF (2626KB) ( 71 )  

Figures and Tables | References | Related Articles | Metrics

In view of the problem of low initial optimization efficiency and poor model accuracy when constructing surrogate models with small sample data, a surrogate-assisted multi-objective particle swarm optimization algorithm based on the combined infill sampling criterion is proposed in this study. The algorithm combines the Kriging model and the radial basis function network model into a heterogeneous ensemble model through the weighted average method, and uses the combined infill sampling criterion of the improved expectation criterion and the minimum surrogate model prediction criterion to manage the surrogate model to speed up the convergence of the model. In addition, the proposed algorithm adopts the actual objective function to evaluate the sample points added in each iteration, and updates the surrogate model to increase the model accuracy. The experimental results show that compared with the non-surrogate model algorithm, the proposed algorithm reduces the evaluation times of the fitness function by 10 times, which proves that the proposed algorithm can improve the optimization efficiency and accuracy of the surrogate model, and achieve a better balance between exploration and development.

Reconfigurable Convolutional Neural Network Accelerator Based on Winograd Algorithm

YUAN Ziang,NI Wei,RAN Jingnan

Electronic Science and Technology. 2022, 35(12):  35-42.  doi:10.16180/j.cnki.issn1007-7820.2022.12.005

Abstract ( 332 )   HTML ( 6 )   PDF (1915KB) ( 49 )  

Figures and Tables | References | Related Articles | Metrics

Neural network is widely used in pattern recognition, predictive analysis, data fitting and other aspects, and it is an important foundation of artificial intelligence. Due to the large calculation amount of convolution and the large amount of network parameters, neural networks have caused problems such as long calculation time and high data access pressure. In response to the above problems, this study accelerates the convolution calculation based on the Winograd algorithm, designs an optimized hardware calculation structure, which improves the data reuse efficiency and calculation parallelism. Compared with the sliding window convolution, this accelerator increases the calculation efficiency by 4.352 times. In terms of convolution kernel gradient calculation, this accelerator adopts an optimized data distribution method, which reduces data movement and meets the data requirements of multiple PE parallel calculations. Compared with the CPU, the performance is improved by 23 times. Experiments show that the convolution calculation throughput rate of the accelerator can reach 192.55 GFLOPS under the VGG-9 network model, and the recognition rate of the CIFAR-10 data set after training is 76.54%.

Multi-Task Traffic Management Algorithm Based on FPGA

ZHANG Fengyin,GAO Bo,JI Yawei

Electronic Science and Technology. 2022, 35(12):  43-48.  doi:10.16180/j.cnki.issn1007-7820.2022.12.006

Abstract ( 192 )   HTML ( 2 )   PDF (1518KB) ( 39 )  

Figures and Tables | References | Related Articles | Metrics

In view of the packet loss problem of data burst transmission in FPGA multi-application tasks, this study proposes a RFCF algorithm for FPGA traffic management. Based on the rate-based traffic management strategy, the receiver feeds back the remaining resources to the sender by controlling the characters. The sender adopts different data priority transmission strategies according to the resource margin, so as to ensure that the burst data can be processed in time under multi-application tasks. The algorithm is implemented and verified in FPGA board. The results show that the RFCF algorithm can effectively improve the data transmission capability of FPGA in multi-application task scenarios. Compared with the rate-based traffic management algorithm, the packet loss rate of the proposed method is reduced by 14.9%.

Servo Control for the Dual Three-Phase Permanent Magnet Synchronous Motor Based on FPGA

YUAN Qingqing,HU Xu,LIU Zhiyong,MA Ting,JIANG Quan

Electronic Science and Technology. 2022, 35(12):  49-56.  doi:10.16180/j.cnki.issn1007-7820.2022.12.007

Abstract ( 665 )   HTML ( 12 )   PDF (3742KB) ( 72 )  

Figures and Tables | References | Related Articles | Metrics

Dual three-phase permanent magnet synchronous motor has been widely used for its reliable performance, high control precision, small output torque ripple and other advantages. In view of the control requirements of high reliability, high precision and small volume of servo motor in aerospace field, this study proposes a servo control algorithm and a specific implementation scheme of dual three-phase permanent magnet synchronous motor based on monolithic FPGA. Based on the hardware platform of Xilinx Kintex7 series XC7K325TFFG900 FPGA chip, the vector control algorithm of dual three-phase three-ring permanent magnet synchronous motor based on double DQ coordinate transformation is realized by Verilog hardware description language. In the verification of the experimental platform, the core control indicators of the current loop bandwidth of 600 Hz and the servo position loop bandwidth of 12 Hz have been a chieved.

Real-Time Image Semantic Segmentation Based on Contextual Attention Mechanism

YU Runrun,JIANG Xiaoyan,ZHU Kaiying,JIANG Guanghao

Electronic Science and Technology. 2022, 35(12):  57-63.  doi:10.16180/j.cnki.issn1007-7820.2022.12.008

Abstract ( 290 )   HTML ( 5 )   PDF (2594KB) ( 55 )  

Figures and Tables | References | Related Articles | Metrics

To address the problem that feature extracted by real time semantic segmentation model lacks contextual information, which causese inconsistent segmentation results between and within classes, a lightweight adaptive spatial attention model and channel attention model are proposed in this study. The adaptive channel attention module uses depthwise separable convolution to model the channel-level feature dependencies, adaptively adjusts the channel convolution kernel size, strengthens the contextual representation ability of high-level features, and enhances the intra-class consistency of segmentation results. The spatial attention module uses grouped convolution to obtain a larger flow area of feature information with a small amount of calculation, strengthens the contextual connection of features at the spatial level, enhances the spatial detail information of features, and enhances the inter-class distinguishability of segmentation results. Testing and analysis on the Cityscapes data set show that the lightweight contextual attention mechanism achieves 71.5% mIoU.

EV Clustering Optimization Community Load Strategy Based on Flexible Optimization Mechanism

DUAN Jundong,HUANG Hongye,WANG Shuaiqiang

Electronic Science and Technology. 2022, 35(12):  64-71.  doi:10.16180/j.cnki.issn1007-7820.2022.12.009

Abstract ( 141 )   HTML ( 1 )   PDF (3267KB) ( 27 )  

Figures and Tables | References | Related Articles | Metrics

In view of the load fluctuation caused by EV disordered charging, this study proposes an EV clustering optimization community load strategy based on flexible optimization mechanism. The strategy divides the community load sequence and EV sequence according to the EV return time, and the EV vehicle sequence is divided into the strategic response vehicle group and the common vehicle group according to the policy responsivity. The grid side and the strategic response vehicle group sign contract A and contract B with flexible limit, and formulate the charging and discharging plan of each vehicle group. Contract A focuses on considering the user benefit, and optimizes the community load sequence of each vehicle sequence to obtain the discharge revenue by controlling the strategic response vehicle group. Contract B focuses on the demand of car usage, and adjusts the charge and discharge plan of EV to balance the contract execution and EV availability, while minimizing the fluctuation of community load sequence. In this study, the household load of a certain community is taken as an example, and the objective function is to minimize the load peak-to-valley difference and the user's expenditure cost, and the contract scenarios are solved by joint modeling through MATLAB, Yalmip platform and Gurobi solver. The results show that the peak valley difference of community load in each contract scenario is reduced by 3.74%, 2.87% and 5.04% respectively after the implementation of the strategy, and the EV sequence cost expense is reduced by 10.80%, 5.23% and 10.55%, respectively.

Visual Question Answer Transmission Attention Network Based on Multi-Modal Fusion

WANG Mao,PENG Yaxiong,LU Anjiang

Electronic Science and Technology. 2022, 35(12):  72-77.  doi:10.16180/j.cnki.issn1007-7820.2022.12.010

Abstract ( 214 )   HTML ( 1 )   PDF (2829KB) ( 32 )  

Figures and Tables | References | Related Articles | Metrics

In view of the shortcomings of traditional visual question answering tasks that cannot fully capture the complex correlation between multi-modal features, this study proposes a visual question-and-answer transmission attention network based on multi-modal fusion. In the feature extraction part, GloVe word embedding + LSTM is used to extract problem features, and ResNet-152 network is adopted to extract image features. Multi-modal fusion is performed through a 3-layer transfer attention network to learn global multi-modal embedding information, which is then used to recalibrate the input features. In addition, a multi-modal transmission attention learning architecture is designed. Through overlapping calculations on the transmission network, the combined features focus on the fine-grained parts of the image and the question, which improves the accuracy of the predicted answer. The experimental results on the VQA v1.0 data set show that the overall accuracy of the model reaches 69.92%, which is improved to varying degrees compared with the accuracy of the other 5 mainstream visual question answering models, indicating the effectiveness of the model and robustness.

K-Anonymity Data Publishing Algorithm Based on Hybrid Clustering

FANG Kai,SHI Zhicai,JIA Yuanyuan

Electronic Science and Technology. 2022, 35(12):  78-83.  doi:10.16180/j.cnki.issn1007-7820.2022.12.011

Abstract ( 134 )   HTML ( 0 )   PDF (804KB) ( 35 )  

Figures and Tables | References | Related Articles | Metrics

In order to reduce the loss of information in data publishing, a k-anonymous data publishing algorithm based on hybrid clustering is proposed to solve the problem of low data availability in existing data anonymity schemes based on clustering. Compared with the traditional single clustering method, the proposed algorithm combines partition clustering and distance clustering, selects the initial clustering center point according to the density characteristics of the data set, and uses partition clustering to achieve the optimal clustering iteratively. In addition, the proposed method eliminates part of the outlier noise in the data set to reduce its impact on the clustering results. For hybrid data records, the distance measurement method combining k-means and k-modes is adopted, and the bucket generalization algorithm is introduced to reduce the information loss caused by generalization operation. Experimental results show that compared with the existing methods, the k-anonymity data publishing algorithm based on hybrid clustering can effectively reduce the information loss of data anonymity and improve the quality of data publishing.

Eight-Section Brocade Sequence Action Recognition and Evaluation Based on Pose Estimation

SU Bo,CHAI Ziqiang,WANG Li,CUI Shuaihua

Electronic Science and Technology. 2022, 35(12):  84-90.  doi:10.16180/j.cnki.issn1007-7820.2022.12.012

Abstract ( 902 )   HTML ( 61 )   PDF (2728KB) ( 205 )  

Figures and Tables | References | Related Articles | Metrics

Action evaluation and feedback can assist fitness exercisers to improve exercise benefits effectively. In order to realize the automatic quantitative evaluation of eight-section brocade movement, a method of recognition and evaluation of human body sequence movements is proposed. The pose estimation algorithm OpenPose is used to extract the coordinates of the key points of the human body, and then normalize them and eliminate redundant points. According to the characteristics of the action, the feature vector of the fusion key points position, distance, joint angle and key points speed is constructed, and the multi-layer perceptron is employed to train the action classification. The accuracy of action recognition on the KTH and self-made eight-section brocade data sets attains to 96.7% and 98.7%, respectively. Based on the eight-section brocade action recognition results, an action sequence is constructed, and the dynamic time warping algorithm is used to calculate the similarity of the two groups of eight-section brocade action sequences. The comparative experimental results show that the similarity can effectively evaluate the integrity and synchronization of actions.

Link Prediction of Knowledge Graph Based on Gaussian Hierarchy-Aware

HU Xueruobai,HUANG Jie,WANG Jiantao,LI Yiming

Electronic Science and Technology. 2022, 35(12):  91-96.  doi:10.16180/j.cnki.issn1007-7820.2022.12.013

Abstract ( 362 )   HTML ( 0 )   PDF (755KB) ( 32 )  

Figures and Tables | References | Related Articles | Metrics

In view of at the problem that the traditional knowledge map link prediction task ignores the possible semantic level between knowledge and the low link prediction results caused by the uncertainty of knowledge, this study proposes a Gaussian level-aware knowledge map link prediction model. In the model, the Gaussian embedding part introduces the Gaussian distribution information of entities and relationships, and the distance between the entity distribution and the relationship distribution is used to measure whether there is a link between entities. The word vector embedding part converts the word vectors of entities and relations into complex vectors. The complex vector of words is mapped to the semantic level of the modeling entities in the polar coordinate system, and the distance between the embedding vectors is used to measure whether there is a link between entities. According to the D-S evidence theory, the score function of the two parts is fused to achieve accurate knowledge map link prediction. The experimental results show that the model can effectively model the semantic level and uncertainty of entities in the knowledge graph, and is superior to other methods on the existing benchmark data sets.

Research on Network Public Opinion Monitoring System Based on Deep Learning

DENG Lei,SUN Peiyang

Electronic Science and Technology. 2022, 35(12):  97-102.  doi:10.16180/j.cnki.issn1007-7820.2022.12.014

Abstract ( 990 )   HTML ( 48 )   PDF (2042KB) ( 206 )  

Figures and Tables | References | Related Articles | Metrics

With the rapid development of the domestic Internet, network public opinion monitoring has become a part of the work of relevant departments and enterprises. Establishment of a public opinion monitoring system can detect public opinion crises in advance and deal with crisis public relations in time. The current study presents a complete framework of network public opinion monitoring system, which consists of four parts: information collection layer, data resource layer, data analysis application layer and application service layer. First, the proposed system can automatically collect data from most portals, microblogs and WeChat accounts, including articles and comments according to keywords. Then, these data are cleaned, segmented and filtered, and the word is embedded using Word2Vec model to obtain the vectorized text. The vectorized text is imported into LSTM deep learning model for sentiment analysis, and the data can be divided into sensitive data, neutral data and non-sensitive data. Finally, the public opinion warning information is displayed by visualization technology. The proposed network public opinion monitoring system can help regulators to monitor and guide relevant public opinions in a timely manner, and promote the harmonious development of society.