Table of Content

07 May 2022, Volume 0 Issue 04

Previous Issue    Next Issue

Data Augmentation for Chinese Named Entity Recognition Task

LI Jian, ZHANG Ke-liang, TANG Liang, XIA Rong-jing, REN Jing-jing

2022, 0(04):  1-6. 

Asbtract ( 518 )   PDF (1025KB) ( 198 )  

References | Related Articles | Metrics

In low-resource natural language processing (NLP) tasks, the existing data is not enough to train an ideal deep learning model. Text data augmentation is an effective method to improve the training effect of such tasks. This paper proposes a group of data augmentation methods based on instance substitution for the task of Chinese named entity recognition. A named entity in the training sample can be replaced by another entity of the same kind without changing the label. The specific algorithms include: 1) crossover substitution between existing entities; 2) synonymous replacement of entity components; 3) automatic generation of Chinese names. These methods are applied to PeopleDailyNER and CLUENER2020 datasets respectively, and the augmentation data is used to train the BERT+CRF model. The experimental results show that the F1 value of the model can be improved by about 10% and 7% respectively on the two datasets with only adding the same amount of augmentation data as the original data under the condition of small samples, and it also has a significant improvement when the training samples increase.

Short-term Load Forecasting of Regional Microgrid Based on LSTM Neural Network

YIN Chun-jie, XIAO Fa-da, LI Peng-fei, ZHAO Qin

2022, 0(04):  7-11. 

Asbtract ( 257 )   PDF (1025KB) ( 153 )  

References | Related Articles | Metrics

There are many studies on the load forecasting of large power grids and relatively few studies on microgrids. Therefore, it is very important to establish a suitable microgrid load forecasting model to improve the accuracy of forecasting. This paper analyzes and selects temperature, daily type, and multiple historical loads as the input variables of the model for the case of fewer input variables, selects the LSTM neural network based on the recurrent neural network for modeling, and constructs the load forecasting model of microgrid based on LSTM neural network. Finally, in order to enhance the reliability of the results, two sets of load data in different time periods are used to predict separately, and the prediction results of the LSTM neural network are compared with those of BP neural network, RBF neural network and Elman neural network. The experimental results show that the prediction results of LSTM neural network are better than BP neural network, RBF neural network and Elman neural network. The LSTM neural network load forecasting model has good promotion prospect under the background of microgrid.





Poverty-returning Prediction Based on Ensemble Learning and Unbalanced Data

GONG Yun-xiang, YUAN Shi-fang, LIU Fu-qian

2022, 0(04):  12-16. 

Asbtract ( 169 )   PDF (1137KB) ( 199 )  

References | Related Articles | Metrics

While China has made the decisive achievement on working on poverty alleviation, there are still some people out of poverty who exist risk of returning to poverty. Based on the unbalanced data set, this paper used the model of SMOTE to do sampling process for multi-class samples of returning to poverty. The sample’s ratio of returned to poverty and non-returned to poverty is 3〖DK〗∶1. After that, based on ensemble learning of Stacking, this paper constructed a prediction model of poverty-returning, used grid search to optimize hyper parameters of every model and improved the generalization ability by combining the 10-fold cross-validation. In this paper, four different integration models are used to predict whether the poor households will return to poverty. Compared with the single model, the experiments indicate that the classification results with fusion model are better. Among them, the optimal Acc and F1-score of fusion model are 0.962 and 0.946.

Prediction of Gearbox Oil Temperature Based on FFT and DNN

ZHEN Chao, TIAN Yu, JI Kun, ZHANG Zheng-kai, HUANG Dao-you

2022, 0(04):  17-20. 

Asbtract ( 150 )   PDF (963KB) ( 73 )  

References | Related Articles | Metrics

Aiming at the non-linearity and correlation of the oil temperature value of the gearbox of wind turbines, in order to achieve accurate oil temperature prediction, a prediction method based on fast Fourier transform (FFT) and deep neural network (DNN) is proposed. First, the time series characteristics of the oil temperature data are analyzed, and the time window is selected to arrange the information. Then, FFT is performed on the information and its high-frequency amplitude characteristics are extracted, and these characteristics are input into the DNN model for training. Finally, an evaluation is made for the output results. The method is validated with measured data, and common models are selected for comparison. The results verify the effectiveness of the method. The method can provide early warning before the gearbox operating state is abnormal, reduce equipment functional failures, and reduce the loss of wind turbines due to failure and shutdown, and has practical value.

PSWGAN-GP: Improved Wasserstein Generative Adversarial Network with Gradient Penalty

CHEN Yun-xiang, WANG Wei, NING Juan, CHEN Yi-dan, ZHAO Yong-xin, ZHOU Qing-hua

2022, 0(04):  21-26. 

Asbtract ( 513 )   PDF (2806KB) ( 115 )  

References | Related Articles | Metrics

The emergence of generative adversarial network (GAN) plays a great role in solving the problem of insufficient sample data in the field of deep learning. In order to solve the detail quality problems of images generated by GAN such as foreground and background separation and contour blurring, this paper proposes an improved Wasserstein generative adversarial network with gradient penalty (PSWGAN-GP) method. Based on the Wasserstein distance loss and gradient penalty of WGAN-GP, this method uses the features extracted from the three pooling layers of the VGG-16 network in the discriminator and calculates the style-loss and perceptual-loss from these features as penalty terms of the original loss, which improves the discriminator’s ability to acquire and discriminate deep features and enhance the details of the generated images. The experimental results show that PSWGAN-GP can effectively improve the quality of generated images with the same generator and discriminator network structure and the same hyperparameters, and the scores in IS and FID are improved relative to other image generation methods.

Gait Feature Recognition Based on Improved Residual Network and Joint Loss Function

HE Xuan, LIU Yi-xin, HE Xiao-hai, QING Lin-bo, CHEN Hong-gang

2022, 0(04):  27-32. 

Asbtract ( 204 )   PDF (1190KB) ( 88 )  

References | Related Articles | Metrics

Aiming at the problems of insufficient recognition accuracy and shallow feature extraction level of the existing gait recognition models, a new joint loss gait feature recognition model Res-GaitSet based on improved residual network is proposed on the basis of GaitSet network. As a unique and effective biometric for long-distance recognition, gait can be widely used in geriatric evaluation, social order security and so on. In the new network, residual elements are introduced into the feature extraction module, and multiple loss functions are used together. This method effectively improves the accuracy and robustness of gait recognition model. The experimental results show that the accuracy of the improved network Res-GaitSet is improved in multiple scenes and different recognition angles of CASIA-B dataset. At the same time, the improved network is used for self built gait data set. Compared with the original network, the recognition effect of the improved network is also improved from different angles, which fully verifies the effectiveness of the improved model.

Multi-target Detection Method Based on Multi-scale CLG Optical Flow Method

REN Chao-yu, ZHAO Dong-e, ZHANG Bin, YANG Xue-feng, CHU Wen-bo

2022, 0(04):  33-37. 

Asbtract ( 171 )   PDF (2761KB) ( 85 )  

References | Related Articles | Metrics

Multi-target detection is an important research direction in computer vision. How to accurately detect targets in military, life, industry and other aspects have very important research significance. Aiming at the poor robustness to noise of traditional optical flow methods in target detection, in order to improve the robustness of the algorithm against noise, the CLG optical flow method is combined with the idea of multi-scale. The main idea of multi-scale is to build an image pyramid, calculate the optical flow vector from coarse to fine, and change the square of L2 norm to L1 norm. Experimental results show that the improved multi-scale CLG optical flow algorithm has better overall performance than the original CLG optical flow algorithm, indicating that the multi-scale CLG optical flow algorithm has better robustness to noise, and can better estimate the optical flow of images.

Wheat Image Recognition Based on Global Self-attention

HE Chen-xi, WANG Zheng-yong, QING Lin-bo, HE Xiao-hai, WU Xiao-qiang

2022, 0(04):  38-44. 

Asbtract ( 187 )   PDF (1460KB) ( 103 )  

References | Related Articles | Metrics

In the actual application scenario, it is very challenging to identify wheat diseases and pests by image recognition. Compared with the previous methods based solely on convolutional neural network (CNN), the method of converting wheat images into a series of visual languages and recognizing wheat from a global perspective is more feasible and practical. The use of convolutional visual Transformers (CVT) to solve wheat recognition is divided into two links. First, two feature maps generated by two-branch CNN are used to realize attentional selective fusion (ASF). ASF obtains different information by fusing multiple features and global-local attention, and projects it into a series of visual languages. Secondly, inspired by the success of Transformers in natural language processing, global self-attention is used to model the relationship between these visual languages. Compared with classical classification networks LeNet-5, ResNet-18, VGG-16 and EfficientNet, CVT improves the recognition rate, and this method has good generalization ability.

Landslide Image Detection Based on Dilated Convolution and Attention Mechanism

LIU Xue-hu, OU Ou, ZHANG Wei-jing, DU Xue-lei

2022, 0(04):  45-51. 

Asbtract ( 300 )   PDF (4767KB) ( 257 )  

References | Related Articles | Metrics

Landslide area image detection and recognition has rich application and research value in disaster scope recognition, disaster data analysis and disaster prevention and mitigation. In this paper, a target detection method combining attention mechanism CBAM and dilated convolution is proposed to solve the problems of the diversity of landslide body shape and texture in landslide image and the unsatisfactory detection and recognition effect of landslide target area. On the basis of the traditional target detection algorithm Faster R-CNN, the attention mechanism model is added to the convolutional neural network layer. The landslide image features are extracted through the CBAM model combining spatial attention and channel attention, and the dilated convolution module is added to enlarge the receptive field area, and to improve the learning ability of the landslide target recognition and non-standard size in the remote sensing image area of the neural network, so as to further improve the detection accuracy of the landslide target area. The experimental results show that, based on the traditional target detection algorithm, the combination of the two methods can improve the recall rate and precision rate of target detection on the remote sensing images of landslides, and it has a certain validity and robustness.





Traffic Sign Recognition Algorithm Based on Improved Residual Network

LIANG Zheng-you, GENG Jing-bang, SUN Yu

2022, 0(04):  52-57. 

Asbtract ( 186 )   PDF (1064KB) ( 97 )  

References | Related Articles | Metrics

For the problems of high-level information loss and insufficient feature extraction in sampling in network structure, the ResNet network structure is improved and a traffic sign recognition method based on multi-scale features and attention mechanism is put forward in this paper. Firstly, multi-scale features are used to fuse different levels of feature information to enrich feature semantic information and enhance the ability of feature extraction. Then, the features of different channels are strengthened through the attention mechanism to improve the overall presence of traffic signs for achieving more accurate traffic sign recognition. The experimental results on GTSRB and BelgiumTS traffic sign datasets show that the accuracies with the proposed methods reach 99.31% and 98.96% respectively, which achieves better results in traffic sign recognition.

Dynamic Load Balancing Strategy of DRC Cluster Based on Nginx

NI Ya-ting, YANG Wen-hui, MIAO Fang, HUANG An-qi, JIANG Yuan

2022, 0(04):  58-64. 

Asbtract ( 205 )   PDF (1301KB) ( 71 )  

References | Related Articles | Metrics

The data-oriented architecture (DOA) provides a new effective solution for the circulation and sharing of massive heterogeneous data. The data registration center (DRC) is the core component of DOA, and its access performance is particularly critical. Aiming at the problem of DRC cluster service overload caused by high concurrent access, Nginx reverse proxy load balancing technology is used to handle high concurrent access. We analyze and optimize the load strategy of Nginx, propose a dynamic load balancing strategy consisting of dynamic configuration, load collection, and algorithm scheduling, and improve the Nginx weighted least connection scheduling algorithm (WLC) in the load scheduling module. The adaptive weight continuously schedules the node with the best performance in the next cycle to process the request. The high concurrency performance test verifies that the proposed load balancing strategy can more effectively deal with the access demand of large traffic in the DRC cluster, improve the resource utilization of the cluster and shorten the request response time.

Analysis on Network Structure of COVID-19’s Review Literature

JIA Fang-di, LIU Ji

2022, 0(04):  65-71. 

Asbtract ( 168 )   PDF (2553KB) ( 61 )  

References | Related Articles | Metrics

The analysis of the network structure of COVID-19’s review literature can provide effective theoretical support for COVID-19’s response to the epidemic situation. The co-citation network analysis of COVID-19’s review literature in Web of Science database is carried out by using visualization software CiteSpace. Through statistics, it is found that the degree distribution of literature co-citation network is power-law distribution, network connectivity is strong and there is a small-world phenomenon. The clique percorlation method is used to mine the overlapping communities, and seven core literatures which span three communities at the same time are detected. Combined with the cited frequency, the function of the network overlapping nodes is verified. With the help of ID symbols and intermediate centrality index of network overlapping nodes, it is found that the corresponding research topics are closely related. Through the important network nodes with different characteristics, we can effectively mine the relevant research topics.

Reconfiguration of Evolutionary Game Network Based on Sparse Bayesian Algorithm

ZHAO Li-na, ZHANG Ya-nan, XIAO Yu-zhu

2022, 0(04):  72-78. 

Asbtract ( 136 )   PDF (1087KB) ( 62 )  

References | Related Articles | Metrics

Evolutionary game is a common type of interaction model in natural and social systems. Exploring the topological structure of an evolutionary game network is the basis for understanding its functions and collective behaviors. For evolutionary game networks, the individual game behavior is usually difficult to be described by dynamic equations, and the related time series information is generally limited and discrete, so it is important to reconstruct the network structure under the limited individual game information. This paper develops the reconstruction method of evolutionary game network based on the sparse Bayesian learning method. The validity of this method is verified by numerical simulation on random networks and small-world networks. Compared with previous L1 norm-based methods, this method can also reconstruct networks with less individual game information, and has higher reconstructing efficiency and stronger noise robustness.

Indoor Positioning Optimization Algorithm Based on Threshold Filtering

CHAI Chen-jing, LIU Bin, PAN Jin-xiao

2022, 0(04):  79-85. 

Asbtract ( 124 )   PDF (1899KB) ( 60 )  

References | Related Articles | Metrics

Aiming at the defects of indoor ultra-wide band (UWB) positioning technology, such as poor positioning effect and inaccurate positioning in complex occlusion environment, this paper proposes a hybrid positioning method based on Chan algorithm and particle swarm optimization algorithm. First, the Chan algorithm is used to obtain the initial estimated position coordinates of the positioning tag, and in a non-line-of-sight (NLOS) environment, a threshold θ is set to filter the position coordinates calculated by the Chan algorithm. The distance difference received by the known base station is summed with the distance difference between different base stations obtained by the tag position information calculated by Chan algorithm. If the sum of the differences is less than the threshold, the position coordinates are directly output. Otherwise, the position coordinates are used as the initial value of the particle swarm algorithm for iteration. Optimization keeps track of individual extreme values and local extreme values, updates individual positions and speeds, and finds the global optimal solution before outputting. The simulation results and the actual field experiment results show that compared with a single algorithm, the hybrid positioning algorithm proposed in this paper improves the positioning accuracy of 27%~31% in the non-line-of-sight environment. The convergence speed is fast, the algorithm complexity is low, and it meets the requirements of indoor positioning.

Non-synchronous OFDM Network Location Technology Based on Multipath Environment

BIAN Zhi-yong, NIU Li-ping

2022, 0(04):  86-91. 

Asbtract ( 114 )   PDF (1140KB) ( 64 )  

References | Related Articles | Metrics

As an advanced communication technology, OFDM plays an important role in the accurate positioning of wireless target node equipment in the wireless base station (BS), but the requirements of signal propagation and arrival time between two wireless devices are relatively high, which makes the application in indoor environment limited. Therefore, based on the known location reference node and BSs transmission, this paper derives the approximate ML position of the target from the collected samples on the target and reference nodes, and proposes an accurate location calculation method of OFDM signal target node based on asynchronous BSs transmission. The proposed method is applied to the existing wireless positioning test system, and the experimental results show that the accurate position calculation method of OFDM signal target node based on asynchronous BSs transmission can accurately locate indoor wireless equipment without any modification of BSs transmission, which has a certain practical significance for the accurate positioning of wireless devices in indoor environment.

EEG Decoding Method Based on Hybrid Feature Selection

MO Yun

2022, 0(04):  92-96. 

Asbtract ( 146 )   PDF (841KB) ( 75 )  

References | Related Articles | Metrics

Motor imagery electroencephalography (EEG) is a multi-channel and high-dimensional signal. Feature selection can reduce the feature dimension and select more discriminative features, thereby effectively improving the performance of EEG decoding. The existing feature selection methods mainly include filter, wrapper and embedded methods, these three methods have their own advantages and disadvantages. In order to comprehensively utilize the advantages of various methods, two hybrid feature selection methods are proposed in this paper. For the first method, the least absolute shrinkage and selection operator (LASSO) is used for feature selection. After the weight of LASSO model is obtained, a series of weight thresholds are set for secondary feature selection. For the second method, the Fisher score is used to score the features, then a series of weight thresholds are set for secondary feature selection. The Fisher linear discriminant analysis (FLDA) is used to classify the feature subsets selected by the two methods. Experiments were conducted on two sets of brain-computer interface (BCI) competition data sets and a set of self-collected laboratory data sets, and the average classification accuracy rates were 77.47%, 76.11%, and 71.30%, respectively. The experimental results show that the classification performance of the proposed method is better than the existing feature selection methods, and the feature selection time also has a greater advantage.

3D Dynamic Visualization Design and Simulation of Wind Field on Web

TIAN Mao-chun, ZOU Xian-yong, YANG Yue, FAN Guang-wei, LAI Hang

2022, 0(04):  97-102. 

Asbtract ( 418 )   PDF (3827KB) ( 94 )  

References | Related Articles | Metrics

In view of the difficulty in displaying the wind field quickly and visually on the Web, a Web-based 3D dynamic wind field visualization design and simulation method was proposed. Firstly, in order to solve the problem that wind field data is difficult to obtain and inconvenient to deal with, the process of wind field data acquisition, conversion, and application was given. Secondly, a method was designed to use Web Worker multi-threading technology to calculate and generate streamlines in parallel to reduce the generation time of wind field streamline tracing and improve the efficiency of streamline generation. Thirdly, combined with color mapping technology, a method to dynamically modify the streamline’s Alpha color channel to characterize the wind field movement was designed. Finally, the WebGL-based visualization engine Cesium was extended to perform the three-dimensional visualization rendering of the wind field, which can directly demonstrate the wind field. The results show that this method not only improves the efficiency of wind field visualization simulation on the Web, but also helps mitigating windstorm disaster.

Audio-visual Eye Fixation Prediction Based on Audio-visual Consistency

YUAN Meng, YU Xiao-yu

2022, 0(04):  103-109. 

Asbtract ( 109 )   PDF (1200KB) ( 63 )  

References | Related Articles | Metrics

The existing audio-visual human eye fixation prediction algorithms use the double-stream structure to extract the features of audio-visual information respectively, and then fuse the audio-visual features to get the final prediction map. However, the audio information and visual information may not be correlated in the datasets. Therefore, when the audio and visual features are inconsistent, the direct fusion of audio and visual features will have a negative impact on the visual features. In view of the above-mentioned problems, this paper proposes an audio-visual consistency network (AVCN) for eye fixation prediction based on audio-visual consistency. In order to verify the reliability of the network, this paper adds an audio-visual consistency network to the existing audio-visual consistency human eye fixation detection model. AVCN carries out the consistency binary judgment on the extracted audio and video features. When the two are consistent, the audio-visual fusion features will be output as the final prediction map; otherwise, the visual dominant features will be output as the final result. The method is tested on six publicly available datasets, and the results show that the proposed AVCN model has better performance.

Virtual Simulation Experiment System for Sponge Campus Renewal Design

ZHANG Jing, MA Xiao-xiao

2022, 0(04):  110-114. 

Asbtract ( 94 )   PDF (1755KB) ( 63 )  

References | Related Articles | Metrics

The existing virtual simulation experiment system does not have an accurate human-computer interaction identification mechanism, which leads to poor signal receiving effect in the process of using virtual simulation equipment. In order to improve the stability of signal link, a virtual simulation experiment system based on sponge campus update is designed and developed. The hardware design of the system is completed by designing the circuit of the single-chip microcomputer system, setting up the coding structure of human-computer interaction, and connecting it to the connecting port of the triode. Through the establishment of campus space rain flood analysis module and sponge campus geographic data visualization module, the modular design of the system is completed. According to the hardware design and software design, the development of sponge campus renewal design virtual simulation experiment system is realized. In the experiment, compared with the last system design, the control rate of total runoff, the utilization rate of rainwater and the reduction rate of runoff pollutants of sponge campus renewal design virtual simulation experiment system are higher.

Improved Image Encryption Algorithm Based on Chaotic Equation and Compressed Sensing Theory

DU Xin-chang, GAO Yu-xiang, CAO Yuan-jie, ZHANG Hao, LIU Hai-bo

2022, 0(04):  115-120. 

Asbtract ( 170 )   PDF (3118KB) ( 68 )  

References | Related Articles | Metrics

In view of the low chaos performance of the traditional Logistic chaotic system and the poor randomness of generated pseudo-random sequences, in this paper, a new improved Logistic chaotic equation is proposed, and a multi-chaotic image compression and encryption system is constructed by combining Lorenz hyperchaotic system and compressed sensing theory. In the process of encryption, the encryption algorithm is combined with the traditional encryption algorithm to obtain the ciphertext image by scrambling and spreading. The controlled measurement matrix is constructed by using the improved Logistic chaotic equation to obtain a pseudo-random sequence with better random performance. Simulation experiments show that the PSNR of the controlled measurement matrix constructed by the improved Logistic chaotic equation reaches 34.26 dB under the condition of 75% compression ratio, which is about 10 dB higher than that of the traditional Logistic chaotic equation under the same conditions. And the algorithm has a good anti-differential attack performance, pixel change rate (NPCR) and uniform mean change degree (UACI) are close to the theoretical value. Therefore, the encryption algorithm proposed in this paper has good compressibility, security and signal reconstruction characteristics.

Malicious TLS Traffic Identification Based on Deep Generation Adversarial Network

QIN Ming-yue, NIAN Mei, ZHANG Jun,

2022, 0(04):  121-126. 

Asbtract ( 326 )   PDF (1151KB) ( 174 )  

References | Related Articles | Metrics

The class imbalance problem in the public data sets of malicious encrypted traffic identification seriously affects the performance of malicious traffic prediction. In this paper, we propose to use the generator and discriminator in the depth generation adversarial network DGAN to simulate the generation of real data sets and the expansion of small sample data to form balanced data sets. In addition, in order to solve the problems that traditional machine learning methods rely on artificial feature extraction, which leads to the decrease of classification accuracy, a malicious traffic recognition model based on the combination of two-way gating loop unit BiGRU and attention mechanism is proposed. The deep learning algorithm automatically obtains the important feature vectors of different time series of data sets to identify malicious traffic. Experiments show that compared with the common malicious traffic recognition algorithms, the model has a good improvement in accuracy, recall, F1 and other indicators, and can effectively realize the identification of malicious encrypted traffic.