Computer and Modernization

Review of Research on Human Behavior Detection Methods Based on Deep Learning

SHEN Jia-wei, LU Yi-ming, CHEN Xiao-yi, QIAN Mei-ling, LU Wei-zhong,

2023, 0(09): 1-9. doi:10.3969/j.issn.1006-2475.2023.09.001

Asbtract ( 406 )

PDF (2112KB) ( 102 )

References | Related Articles | Metrics

Human behavior recognition has always been a hot topic of research in the field of computer vision and video understanding and is widely used in other areas such as intelligent video surveillance and human-computer interaction in smart homes. While traditional human behavior detection algorithms have the disadvantages of relying on too many data samples and being susceptible to environmental noise， evolving deep learning techniques are gradually showing their advantages and can be a good solution to these problems. Based on this， this paper firstly introduces some commonly used behavioral recognition datasets and analyses the current research status of human behavioral recognition based on deep learning， then describes the basic process of behavioral recognition and commonly used behavioral recognition methods， finally summarizes the performance， existing problems of various existing behavioral recognition methods， and outlooks the future development directions.

Fault Diagnosis of Hydraulic Systems Based on CNN-BiLSTM

LIU Fu-qi, ZHANG Da, SONG Jian-hua, WANG Hai-dong

2023, 0(09): 10-19. doi:10.3969/j.issn.1006-2475.2023.09.002

Asbtract ( 93 )

PDF (8241KB) ( 46 )

References | Related Articles | Metrics

Aiming at the fault diagnosis problem of the main components in complex hydraulic system， a fault diagnosis model based on one-dimensional convolutional neural network （1D-CNN） and bidirectional long-term memory network （BiLSTM） is proposed to achieve multi-sensor information fusion and make fault diagnosis of piston pump and throttle valve. In the proposed model， the signals collected by various sensors are carried out data-stage fusion firstly， then the fault characteristics of the fusion signal are extracted by CNN and dimensionality reduction is performed， and then the forward and reverse data characteristics in the signal are learned by BiLSTM， finally the Softmax is used for classification， which realizes the diagnosis of piston pump and throttle valve fault. The experimental results show that the proposed method can automatically extract the fault characteristics in the signal and consider the positive and negative data characteristics contained in the signal. The diagnostic accuracy of the plunger pump can reach 96.3%， and the diagnostic accuracy of the throttle valve can reach 94.28%， which realizes the accurate and reliable diagnosis of the fault state of the plunger pump and the throttle valve.

Personnel Safety Warning System in Industrial Plant Based on Computer Vision

GU Cheng-wei, DING Yong, LI Deng-hua

2023, 0(09): 20-26. doi:10.3969/j.issn.1006-2475.2023.09.003

Asbtract ( 76 )

PDF (2730KB) ( 53 )

References | Related Articles | Metrics

In view of the frequent safety accidents of hoisting machinery in industrial plants， this paper proposes a personnel safety alert system in industrial plants based on computer vision， which uses a combination of computing platform and target detection algorithm to detect the personnel targets in the field operation monitoring video in real time and output corresponding control instructions. The target detection algorithm is based on YOLOv5 network， and the attention mechanism is embedded in the network structure. The space and channel based hybrid attention mechanism module is added to BottleneckCSP module， which can improve the accuracy of small target detection. In addition， a person tracking algorithm is introduced to modify and fuse the detection results， which can reduce the missed detection rate when the person is in the occlusion situation. The improved algorithm is tested in the self built dataset. Compared with the original YOLOv5 network， the improved algorithm is 3.414 percentage point higher on the mAP， and the detection speed can reach 40.3 FPS， which has a good detection effect. Finally， the algorithm model is deployed to the computing platform， and is built and tested on the scene. The test statistics showe that the detection accuracy of ordinary personnel and navigators is 94.4% and 95.1%， respectively， which has good detection performance and can stably perform corresponding automatic security alert operations.

Clustering Method of Cloud Platform Abnormal Transmission Data Based on Hilbert Similarity

WANG Hong-jie, XU Sheng-chao

2023, 0(09): 27-31. doi:10.3969/j.issn.1006-2475.2023.09.004

Asbtract ( 56 )

PDF (1641KB) ( 35 )

References | Related Articles | Metrics

The difference of abnormal transmission data of cloud platform is very small， which leads to poor accuracy of clustering of abnormal transmission data of cloud platform. Therefore， this paper proposes a clustering method of abnormal transmission data of cloud platform based on Hilbert similarity. The proposed method collects the abnormal transmission data of the cloud platform， maps the collected data into the Hilbert space， and constructs the Hilbert index to obtain the discrete probability distribution of indexs. Wavelet basis is used to analyze the sensitivity of abnormal transmission data of cloud platform， and the wavelet basis with relatively low sensitivity is selected for wavelet decomposition of abnormal transmission data. The proposed method calculates the similarity value in Hilbert space， divides it into a same data set， accurately divides the cloud platform abnormal transmission data， and realizes the cloud platform abnormal transmission data clustering. The experimental results show that the number of correct clustering data of the proposed method is 97 groups， and the time for clustering abnormal transmission data is only 146 s， which can effectively distinguish the abnormal transmission data in the cloud platform.

Personalized Recommendation Method of Web Service Resources Based on Maximum Entropy

YANG Liu-qing, WANG Chong

2023, 0(09): 32-37. doi:10.3969/j.issn.1006-2475.2023.09.005

Asbtract ( 39 )

PDF (1541KB) ( 50 )

References | Related Articles | Metrics

When recommending Web service resources based on user behavior characteristics， insufficient consideration of the correlation between different characteristics makes the low F-Measure value of the recommendation method. Therefore， a personalized recommendation method of Web service resources based on maximum entropy is proposed. According to the historical operation records of users， the implicit behavior characteristics of users are extracted from three aspects of user characteristics， commodity characteristics and interaction characteristics to improve the missing information of users. Collaborative filtering method is used to mine the association between users and resources and generate user interest matrix. Based on the principle of maximum entropy calculation， the feature function is constructed to analyze the correlation between features， and based on this， the Web service resource selection algorithm is designed. Finally， the constraints are established according to the basic attributes of users and the resource scoring matrix， and the optimal personalized resource recommendation results are obtained. The experimental results show that compared with the recommendation method based on ontology reasoning and intelligent computing， the F-measure value is increased by 41 percentage points and 33 percentage points， and the resource recommendation results can better meet the needs of users.

Calculation Method of Chaohu Lake Surface Rainfall Based on Ensemble Learning

WANG Jie, XU Xiang, LUO Xiao-dan, ZHANG Meng, HUANG Che, HONG Guan-zhong, WANG Xiang

2023, 0(09): 38-43. doi:10.3969/j.issn.1006-2475.2023.09.006

Asbtract ( 66 )

PDF (2885KB) ( 21 )

References | Related Articles | Metrics

In view of the problems such as insufficient precipitation data due to the difficulties in deployment of the lake meteorological observation station and the complexity of rainfall calculation under the traditional numerical model， this paper takes Chaoahu Lake as the research object， and creates a model dataset using radar 3D mosaic data and precipitation data from meteorological observation station in the Chaohu Lake basin. Then a quantitative precipitation estimation （QPE） model based on ensemble learning is constructed， and the self-made data set is used to train the model. Combined with geographic information system （GIS）， the Chaohu Lake basin is divided by latitude and longitude grid and superimposed on the radar mosaic spatially. The arithmetic mean of the rainfall at each grid point which is calculated by the QPE model is calculated to obtain the rainfall on the lake surface. In the study， the performance of the QPE model based on ensemble learning of random forest （RF）， XGBoost and LightGBM algorithms is compared and analyzed， and the QPE model with better performance is selected to conduct hyperparameter tuning. And the results of lake surface rainfall obtained by grid averaging method and CMA land data assimilation system （CLDAS） are comparatively analyzed. The results show that the QPE model using RF algorithm has better performance， and the overall trend is consistent although there are differences in the results values calculated by the grid average method and CLDAS. In conclusion， this method could be used to calculate the surface rainfall of Chaohu Lake， and provides an important reference for flood prevention and control of Chaohu Lake and its watershed.

Prediction of Bayesian Optimized Gradient Boosting Tree for Interior Natural Illuminance Distribution

JI Xin-cheng, WANG Yan-kai, ZHANG Ying, XU Yan-jie

2023, 0(09): 44-50. doi:10.3969/j.issn.1006-2475.2023.09.007

Asbtract ( 62 )

PDF (4661KB) ( 26 )

References | Related Articles | Metrics

Large prediction errors in lighting models are caused by the nonlinear temporal variation and non-uniform spatial distribution of the natural light that enters the room through the window. In the case of limited data， how to achieve high-precision modeling of interior light environment under natural light is a huge challenge. To solve these problems， an interior illuminance prediction technique using principal component analysis and Bayesian optimized gradient boosting regression tree is proposed. Firstly， the algorithm performs feature reshaping by principal component analysis， fully takes into account the intrinsic correlation between various illuminance data features and preprocesses the sample data using Dummy variable. Then the random forest is used to determine the initial parameters of GBRT to improve its convergence speed and stability. Finally， cross-validation and Bayesian optimization algorithm are integrated to determine the hyperparameter combination of GBRT， so as to further improve the prediction ability of the model for indoor illumination distribution. The experimental results show that under different weather and time conditions， the R2， MAE and RMSE of 600 test samples with illuminance are 0.9912， 18 lx and 40 lx， respectively， which are superior to other algorithms and can significantly reduce sample deviation value.

Prediction of Time Series with Missing Value Based on Tensor Autoregressive Completion

LIU Rui-xue, LI Wen, LIU Fang, DU Shou-guo

2023, 0(09): 51-58. doi:10.3969/j.issn.1006-2475.2023.09.008

Asbtract ( 57 )

PDF (1107KB) ( 33 )

References | Related Articles | Metrics

To solve the prediction problem of high-dimensional time series with missing values， a tensor autoregressive completion algorithm is proposed. Based on the high-precision low-rank tensor completion algorithm （HaLRTC）， the tensor autoregressive norm is added， and the missing data of the tensor time series is completed by making full use of the information of all dimensions of the high-dimensional time series， in which the tensor kernel norm captures the long-term trend of the time series， and the tensor autoregressive norm captures the short-term trend of the time series. Using high-order form of the autoregressive model， the completed high-dimensional time series is predicted. To verify the effectiveness of the algorithm， the core autoregressive tensor completion （CCAR）， the core tensor autoregressive completion （CTAR）， and the tensor core autoregressive completion （TCAR） based on Tucker decomposition are proposed for ablation experiment. The results of ablation experiments and comparison experiments with other existing methods show that the proposed algorithm has obvious prediction advantages in the case of small proportion of missing data.

Point Cloud Registration Algorithm Based on Combined Feature Points and#br# Principal Component Analysis#br#

ZHANG Ya-wen, LIN Wen-zhong, HAN Xiao-dong

2023, 0(09): 59-63. doi:10.3969/j.issn.1006-2475.2023.09.009

Asbtract ( 69 )

PDF (1711KB) ( 36 )

References | Related Articles | Metrics

Aiming at the problems of low accuracy， easy mismatching， and the descriptive error of single point features in the subsequent series of improved algorithms to point cloud shape， a point cloud registration algorithm based on point cloud combination feature point and principal component analysis is proposed. The intrinsic shape signatures is extracted from the point cloud， and the AC algorithm is used to extract the boundary points（BDRY） of the point cloud to form the combined feature points （ISS_BDRY）. The normal of the ISS_BDRY feature point is calculated and described by fast point feature histogram， and then the sampling consistency initial registration algorithm improved by principal component analysis SAC-IA is used to minimize the distance error between the main axes of the point cloud， thereby reducing the number of iterations in the point cloud fine registration process， and providing good pose for subsequent point cloud registration. In the fine registration stage， the iterative closest point registration algorithm introduced KD-Tree to accelerate search point cloud is used for registration. The experimental results show that compared with other single-point features， the registration accuracy of extracted combined feature points on Cat and Michael point clouds reaches 10-8 orders of magnitude， and the registration accuracy of the combined feature method is increased by 65.19% and 44.77%， respectively. Compared with ICP， NDT， Super 4PCS and other algorithms， the accuracy of the fine registration stage reaches 10-16 orders of magnitude， and it is almost completely coincide.

SDN Reliability Evaluation Algorithm Based on Improved BDD

JIANG Hou-hai, ZHUANG Yi, CAO Zi-ning

2023, 0(09): 64-69. doi:10.3969/j.issn.1006-2475.2023.09.010

Asbtract ( 50 )

PDF (1495KB) ( 21 )

References | Related Articles | Metrics

Aiming at the problem of insufficient reliability analysis of SDN data forwarding layer， this paper proposes an SDN reliability evaluation algorithm based on BDD， which can complete fast and accurate reliability analysis of SDN data forwarding layer. In view of the shortcomings of traditional BDD ranking methods that lead to large scale and long construction time for network reliability evaluation models， this paper proposes a new heuristic edge ranking algorithm， MP-BFS， to rank the variables of BDD. The experimental results show that compared with the traditional sorting algorithm， the proposed MP-BFS algorithm can significantly reduce the construction scale of BDD and complete the construction of BDD faster. The BDD-SDN algorithm can be used for rapid and accurate reliability analysis on the forwarding layer of SDN data.

SFIM Image Fusion Method Combining IHS and Adaptive Filtering

TANG Yu-lin, HUANG Deng-shan, CHEN Shu-lu, CHEN Peng-ming

2023, 0(09): 70-76. doi:10.3969/j.issn.1006-2475.2023.09.011

Asbtract ( 67 )

PDF (3327KB) ( 30 )

References | Related Articles | Metrics

SFIM is a commonly used panchromatic-multispectral fusion algorithm with good spectral injection capability， but the poor quality of spatial information incorporation is due to the poor acquisition of ideal low-resolution panchromatic images. Aiming at the shortcomings of SFIM， a SFIM model combining IHS and Gaussian filtering is proposed. The method uses an adaptive linear combination to obtain the I-Component of the multispectral image. Then， with the average gradient of the best adjusted I-component as the standard， the ideal low-resolution panchromatic image is determined by Gaussian filtering of the downsampled panchromatic image. Finally， the multispectral image and the low-resolution panchromatic image are upsampled to the same size as the panchromatic image， and the SFIM transform is performed to obtain the fusion results. The experiments are carried out on the data of GF-2 and ZY3-1. The experimental results show that the algorithm better overcomes the shortcomings of IHS and SFIM and better performs in both qualitative and quantitative analysis. It has better spectral retention and injects more detailed spatial detail information， which effectively improves the quality of fused image detail information. This experiment can provide a useful value reference for the study of panchromatic-multispectral image fusion.

Formal Verification of Raft Protocol Based on Probabilistic Model

GUAN Jin-ping, YANG Jin-ji, YANG Cheng-long

2023, 0(09): 77-81. doi:10.3969/j.issn.1006-2475.2023.09.012

Asbtract ( 53 )

PDF (2230KB) ( 30 )

References | Related Articles | Metrics

Consensus protocol is important element and core component of distributed system， it is dedicated to solve the problem of ensuring the same data consistency among nodes that may fail in a distributed scenario， and its accuracy and efficiency directly determine the performance of the system. Raft consensus protocol is common and effective consensus algorithm in current distributed systems. This paper firstly models the Raft consensus protocol formally using a probabilistic model detection method， then describes its relevant properties using property spcification technique of probabilistic model checking， and finally verifies and analyses the consistency and efficiency of the Raft consensus protocol. The experimental results show that it satisfies consistency， but the number of election rounds increases during the leader election phase when the difference range of the latest log serial numbers maintained by followers increases， resulting in the election time increase throughout the service cycle， thus affecting the execution efficiency of Raft.

Image Classification Based on Deep Feedback CNN

WU Tian, LIU Hai-hua, TONG Shun-yan

2023, 0(09): 82-86. doi:10.3969/j.issn.1006-2475.2023.09.013

Asbtract ( 65 )

PDF (2093KB) ( 32 )

References | Related Articles | Metrics

For image classification processing， convolutional neural network （CNN） is a common method. But the current methods based on CNN construction do not make full use of the perceptual characteristics of visual neurons， so that the network loses a lot of important image feature information in the process of learning. Therefore， starting from the perceptual characteristics of visual neurons， this paper proposes a deep feedback convolutional neural network model that conforms to visual perception. In this model， the feedback regulation mechanism of visual neurons is simulated， and the deep feedback recurrent neural network （DF-RNN） is constructed. At the same time， combining the advantages of DF-RNN and CNN， DF-RNN is embedded in CNN to exert its associative memory function， and then deep features are extracted from shallow features through DF-RNN. In addition， because the weight parameters of DF-RNN adopt a sharing mechanism， the number of parameters for network training is greatly reduced. Finally， the image classification experiment on the Oxford flowers-102 standard dataset is carried out by the network model， and the classification accuracy can reach 86.8%， which is 9.6 percentage points higher than VGG16. It shows the effectiveness of the proposed network model.

Three-dimensional Spatial Location Method for Inter-locking Holes of Fracture Bones Based on Deep Regression

WANG Fei, JIANG Jun-feng, DENG Zi-yue, CHEN Liang, HUANG Rui, YAO Qing-qiang

2023, 0(09): 87-93. doi:10.3969/j.issn.1006-2475.2023.09.014

Asbtract ( 41 )

PDF (3476KB) ( 19 )

References | Related Articles | Metrics

After the fracture of the human long bone， the intramedullary nail needs to be inserted into the medullary cavity to fix the fracture area. Then the screw is required to be placed in the distal hole of the nail. The difficulty of this operation is that the distal hole’s pose will change with the shape of the medullary cavity. Therefore， it is difficult to accurately locate the distal hole’s pose by the traditional manual method. To solve this problem， a three-dimensional spatial location method using single intraoperative X-ray image for distal hole is proposed. Firstly， the initial pose of the distal hole is determined by comparing the contour similarity between the virtual and real nails. Then， the nail’s contour is aligned using a correction algorithm to drive the iterative optimization of the virtual nail pose. Finally， the accurate pose of distal hole is determined. In addition， in order to extract the contour features of X-ray film accurately， a two-stage extraction method from coarse to fine intramedullary nail contour is proposed combined with the target detection algorithm. The experiments are carried out in simulated and clinical environments. Comparison is drawn between the calculated distal hole’s axis and the real hole’s axis. The distance and angle errors in the simulated environment are 0.42 mm， 0.46°， and the errors in the clinical environment are 0.75 mm， 0.81°. The proposed method can meet the actual surgical needs and improves the efficiency and planning in the distal locking nail surgery.

COVID-19 X-ray Classification Based on Improved Efficientnet Network

LIU Chan-yi, HUANG Dan, XUE Lin-yan, WANG Tao, ZHU Tao,

2023, 0(09): 94-99. doi:10.3969/j.issn.1006-2475.2023.09.015

Asbtract ( 54 )

PDF (1972KB) ( 45 )

References | Related Articles | Metrics

In response to many problems such as the rapid spread of new coronary pneumonia， the time-consuming process of manual diagnosis using COVID-19 medical images， the imbalance of medical resources and the pressure of doctors’ diagnosis， this paper introduces a new attention module ECBAM on the basis of the lightweight network EfficientNet-B0 and proposes the EfficientNet-ECBAM network. Firstly， replacing the SE module in the EfficientNet-B0 network structure with this module can improve the problem that some details of the SE module are lost due to the downscaling operation. Secondly， because the ECBAM module can extract features in both channel and space dimensions， it can also improve the problem that the SE module extracts insufficient information of image features. On the selected COVID-19 chest X-ray dataset， compared with the classical convolutional neural classification network VGG16 and ResNet-50， the accuracy of the improved EfficientNet-ECBAM network based on the EfficientNet-B0 network is improved by 3.76 percentage points and 2.13 percentage points respectively， specificity and sensitivity are also improved. The number of model parameters is also reduced by 97.3% and 85.6% respectively. Compared with the lightweight network SqueezeNet and MobileNet V1， the accuracy of EfficientNet-ECBAM is improved by 2.97 percentage points and 2.44 percentage points respectively. The improved ECBAM module also outperforms other attention modules in the ablation experiments in all metrics. The experimental results show that the EfficientNet-ECBAM network model proposed in this paper has the advantages of good classification performance， low number of parameters and low computation， which is favorable for deployment in medical institutions in less economically developed areas.

Ultrasonic Image Diagnosis of Hepatic Echinococcosis Based on Deep DenseNet Network

MA Guo-xiang, YANG Ling-fei, YAN Chuan-bo, ZHANG Zhi-hao, SUN Bing, WANG Xiao-rong

2023, 0(09): 100-104. doi:10.3969/j.issn.1006-2475.2023.09.016

Asbtract ( 70 )

PDF (2099KB) ( 50 )

References | Related Articles | Metrics

Hepatic echinococcosis is a serious regional parasitic disease. The diagnosis and classification of lesions mainly rely on the subjective judgment of the clinician on ultrasound images. In areas with weak medical conditions， the screening and diagnosis of the disease can easily be misjudged. In order to improve the diagnosis efficiency and accuracy of liver hydatid disease， this paper combines deep learning algorithms to apply the deep DenseNet network to the image classification problem of liver hydatid disease， and uses the powerful feature extraction capabilities of deep convolutional neural networks to construct liver hydatid classification model. In addition， in order to be able to provide more reliable image input information， the ROI-based preprocessing method is used to extract the lesion ROI area of the original image. Finally， the model is evaluated and verified on the constructed data set， and the accuracy can reach 93%， and by using gradient weighted class activation map for visual analysis， it showes that the model has strong robustness and better classification effect.

RFID System Security Analysis Model Based on Stochastic Petri Net

XIAO Hang, LI Peng, MA Hui-ping, ZHU Feng,

2023, 0(09): 105-114. doi:10.3969/j.issn.1006-2475.2023.09.017

Asbtract ( 28 )

PDF (7459KB) ( 17 )

References | Related Articles | Metrics

To solve the problem of RFID system breakdown risk caused by frequent RFID system attacks， this paper proposes a RFID system security analysis model based on hierarchical generalized stochastic Petri net. The model uses the existing knowledge reserve to simulate the real RFID virtual environment， accurately and effectively deduces the attack process in the RFID system， and quantifies the risk of the RFID system. Firstly， the RFID attacker model is constructed using the information of attack hierarchy，attack authority and permission-based attacks. Secondly， the description of the attacker's behavior is modeled and described its impact on the RFID system state. Finally， based on the constructed model， the attack probability， weak nodes and other aspects of the RFID system are assessed. The experimental results show that the proposed model can effectively evaluate the risk of RFID system， and greatly reduce the complexity of evaluation time.

Continuous Attribute Discretization Algorithm of Rough Sets for BP Neural Networks Based on Particle Swarm Optimization#br#

MAO Ming-yang, XU Sheng-chao

2023, 0(09): 115-119. doi:10.3969/j.issn.1006-2475.2023.09.018

Asbtract ( 37 )

PDF (1410KB) ( 28 )

References | Related Articles | Metrics

When discretizing the continuous attributes of rough sets， the obtained breakpoint set is not the optimal set， resulting in poor discretization effect. Therefore， a particle swarm optimization BP neural network oriented discretization algorithm for continuous attributes of rough sets is proposed. The discretization of continuous attributes of rough sets is analyzed. Particle swarm optimization algorithm is used to improve the weights and thresholds in BP neural network. Based on the optimized BP deep neural network， information systems with continuous attributes are classified， so as to obtain multiple breakpoints， form sub-breakpoint sets， and build candidate breakpoint sets. The candidate breakpoint set is mapped to particles in particle swarm optimization algorithm， and the best breakpoint set is found by changing the speed and position of particles to complete the discretization of continuous attributes of rough set. The experiment results show that the proposed method can better realize the discretization of continuous attributes， and the data consistency is close to 100% when it is the highest， and the classification accuracy and convergence speed are higher under different algorithms， which shows that this method has strong application prospect.

Adaptive Protection System for Distribution Network Considering Grid Connection of Distributed Generation

LIU Xian-zhuo, DENG Wei-si, XIE En-yan

2023, 0(09): 120-126. doi:10.3969/j.issn.1006-2475.2023.09.019

Asbtract ( 32 )

PDF (1387KB) ( 46 )

References | Related Articles | Metrics

To further improve the rapidness of adaptive protection of distribution network considering distributed generation grid connection， an adaptive protection system for distribution network considering distributed generation grid connection is proposed. The improved microgenetic algorithm is utilized to optimize and update the settings of protective relays in real time. In order to verify the effectiveness of the proposed adaptive protection system， a simulation database is built by Simulink/MATLAB platform for the simulation of multiple fault cases. The effect of the proposed adaptive protection system on fault line removal in the simulation database is compared with that of other two adaptive protection systems. The comparison results show that the fault line removal speeds obtained by the proposed adaptive protection system are significantly faster than those obtained by the other two adaptive protection systems.

Table of Content