Table of Content

10 January 2023, Volume 43 Issue 1

Previous Issue    Next Issue

Artificial intelligence

Federated learning algorithm for communication cost optimization

ZHENG Sai, LI Tianrui, HUANG Wei

2023, 43(1):  1-7.  DOI: 10.11772/j.issn.1001-9081.2021122054

Asbtract ( )   HTML ( )   PDF (1156KB) ( )  

References | Related Articles | Metrics

Federated Learning （FL） is a machine learning setting that can protect data privacy， however， the problems of high communication cost and client heterogeneity hinder the large?scale implementation of federated learning. To solve these two problems， a federated learning algorithm for communication cost optimization was proposed. First， the generative models from the clients were received and simulated data were generated by the server. Then， the simulated data were used by the server to train the global model and send it to the clients， and the final models were obtained by the clients through fine?tuning the global model. In the proposed algorithm only one round of communication between clients and the server was needed， and the fine?tuning of the client models was used to solve the problem of client heterogeneity. When the number of clients is 20， experiments were carried out on MNIST and CIFAR?10 dataset. The results show that the proposed algorithm can reduce the amount of communication data to 1/10 of that of Federated Averaging （FedAvg） algorithm on the MNIST dataset， and can reduce the amount of communication data to 1/100 of that of Federated Averaging （FedAvg） algorithm on the CIFAR-10 dataset with the premise of ensuring accuracy.

Graph representation learning by autoencoder with one-shot aggregation

YUAN Lining, LIU Zhao

2023, 43(1):  8-14.  DOI: 10.11772/j.issn.1001-9081.2021101860

Asbtract ( )   HTML ( )   PDF (2556KB) ( )  

References | Related Articles | Metrics

AutoEncoder （AE） is an efficient learning model for graph data representation， but most of the Graph AutoEncoders （GAEs） are shallow models， their efficiencies decrease with the increase of hidden layers. Aiming at the above problems， a new GAE model OSA-GAE and a new Variational Graph AutoEncoder OSA-VGAE were proposed based on One-Shot Aggregation （OSA） and Exponential Linear Unit （ELU）. Firstly， the encoder was constructed by a multi-layer Graph Convolutional Network （GCN）， and OSA and ELU function were introduced. Then， the topology of the graph was recovered by the inner product decoder in the decoding stage. In addition， a regularization term was introduced to the loss function in order to prevent parameter overfitting during the model training process. Experimental results show that OSA and ELU function can improve the performance and gradient information transmission of the deep GAEs. In the link prediction task of benchmark citation dataset PubMed， when using 6-layer GCN， deep OSA-VGAE improves Area Under ROC Curve （AUC） and Average Precision （AP） by 8.67 and 6.85 percentage points respectively compared to the original VGAE， and deep OSA-GAE improves AP and AUC by 6.82 and 4.39 percentage points respectively compared to the original GAE.

Multi-round conversational reinforcement learning recommendation algorithm via multi-granularity feedback

YAO Huayong, YE Dongyi, CHEN Zhaojiong

2023, 43(1):  15-21.  DOI: 10.11772/j.issn.1001-9081.2021111875

Asbtract ( )   HTML ( )   PDF (1249KB) ( )  

References | Related Articles | Metrics

Multi-round Conversational Recommendation System （CRS） obtains real-time information of users interactively， thus performing better than traditional recommendation methods such as collaborative filtering based method. However， existing CRS suffers from problems inaccurate mining of user preferences， too many conversational rounds required and inappropriate recommendation moments. Aiming at these problems， a new conversational recommendation algorithm based on deep reinforcement learning considering user’s multi-granularity feedback information was proposed. Different from existing CRS， in each conversation， the feedback of users on items themselves and more fine-grained item attributes was considered by the proposed algorithm at the same time. Then， users， items and attribute features of items were updated online by using the collected multi-granularity feedback， and the environment state after each round of conversation was analyzed by Deep Q-Network （DQN） algorithm. As a result， more appropriate and reasonable decisions were made by the system， and the reasons of why user buying items were analyzed and the users’ real-time preferences were mined comprehensively with fewer conversation rounds. Experimental results on two real datasets show that compared with Simple Conversational Path Reasoning （SCPR） algorithm， the proposed algorithm has the 15 turns success rate increased by 46.5%， and the 15 average turns decreased by 0.314 rounds in Last.fm dataset， while it maintains the same level of success rate but the 15 average turns decreased by 0.51 rounds in Yelp dataset.

Novelty detection method based on dual autoencoders and Transformer network

ZHOU Jiahang, XING Hongjie

2023, 43(1):  22-29.  DOI: 10.11772/j.issn.1001-9081.2021111983

Asbtract ( )   HTML ( )   PDF (2078KB) ( )  

References | Related Articles | Metrics

AutoEncoder （AE） based novelty detection method utilizes reconstruction error to classify the test samples to be normal or novel data. However， the above method produces very close reconstruction errors on normal data and novel data. Therefore， some novel data are easy to be misclassified as normal data. To solve the above problem， a novelty detection method composed of two parallel AEs and one Transformer network was proposed， namely Novelty Detection based on Dual Autoencoders and Transformer Network （DATN-ND）. Firstly， the bottleneck features of input samples were used by Transformer network to generate the bottleneck features with pseudo-novel data， thereby increasing the novel data information in the training set. Secondly， the bottleneck features with novel data information were reconstructed by the dual AEs to normal data as much as possible， increasing the reconstruction error difference between novel and normal data. Compared with MemAE （Memory-augmented AE）， DATN-ND has the Area Under the Receiver Operating Characteristic curve （AUC） improved by 6.8 percentage points， 12.0 percentage points， and 2.5 percentage points respectively on MNIST， Fashion-MNIST， and CIFAR-10 datasets. Experimental results show that DATN-ND can effectively make the difference of reconstruction error between normal data and abnormal data bigger.

Unlabeled network pruning algorithm based on Bayesian optimization

GAO Yuanyuan, YU Zhenhua, DU Fang, SONG Lijuan

2023, 43(1):  30-36.  DOI: 10.11772/j.issn.1001-9081.2021112020

Asbtract ( )   HTML ( )   PDF (1391KB) ( )  

References | Related Articles | Metrics

To deal with too many parameters and too much computation in Deep Neural Networks （DNNs）， an unlabeled neural network pruning algorithm based on Bayesian optimization was proposed. Firstly， based on a global pruning strategy， the sub-optimal compression ratio of the model caused by layer-by-layer pruning was avoided effectively. Secondly， the pruning process was independent on the labels of data samples， and the compression ratios of all layers were optimized by minimizing the distance between the output features of pruning and baseline networks. Finally， the Bayesian optimization algorithm was adopted to find the optimal compression ratio of each layer， thereby improving the efficiency and accuracy of sub-network search. Experimental results show that when compressing VGG-16 network by the proposed algorithm on CIFAR-10 dataset， the parameter compression ratio is 85.32%， and the Floating Point of Operations （FLOPS） compression ratio is 69.20% with only 0.43% accuracy loss. Therefore， the DNN model can be compressed effectively by the proposed algorithm， and the compressed model can still maintain good accuracy.

Aspect-based sentiment analysis model embedding different neighborhood representations

LIU Huan, DOU Quansheng

2023, 43(1):  37-44.  DOI: 10.11772/j.issn.1001-9081.2021122099

Asbtract ( )   HTML ( )   PDF (1680KB) ( )  

References | Related Articles | Metrics

The Aspect-Based Sentiment Analysis （ABSA） task aims to identify the sentiment polarity of a specific aspect. However， the existing related models lack the short-distance constraints on the context of the aspect word for the natural sentences with uncertain structure， and easily ignore the syntactic relations， so it is difficult to accurately determine the sentiment polarity of the aspect. Aiming at the above problems， an ABSA model with Embedding Different Neighborhood Representations （EDNR） was proposed. In this model， on the basis of obtaining the word order information of sentences， the nearest neighbor strategy combining with Convolution Neural Network （CNN） was used to obtain aspect neighborhood information， so as to reduce the influence of far irrelevant information on the model. At the same time， the grammatical information of sentences was introduced to increase the dependency between words. After fusing the two features， Mask and attention mechanism were used to pay special attention to the aspect information and reduce the interference of useless information to the sentiment analysis model. Besides， in order to evaluate the influence degree of contextual and grammatical information on sentiment polarity， an information evaluation coefficient was proposed. Experiments were carried out on five public datasets， and the results show that compared with the sentiment analysis model AGCN-MAX （Aggregated Graph Convolutional Network-MAX）， the EDNR model has the accuracy and F1 score on dataset 14Lap improved by 2.47 percentage points and 2.83 percentage points respectively. It can be seen that the EDNR model can effectively capture emotional features and improve the classification performance.

Aspect-based sentiment analysis model integrating match-LSTM network and grammatical distance

LIU Hui, MA Xiang, ZHANG Linyu, HE Rujin

2023, 43(1):  45-50.  DOI: 10.11772/j.issn.1001-9081.2021111874

Asbtract ( )   HTML ( )   PDF (1828KB) ( )  

References | Related Articles | Metrics

Aiming at the problems of the mismatch between aspect words and irrelevant context and the lack of grammatical level features in Aspect-Based Sentiment Analysis （ABSA） at current stage， an improved ABSA model integrating match-Long Short-Term Memory （mLSTM） and grammatical distances was proposed， namely mLSTM-GCN. Firstly， the correlation between the aspect word and the context was calculated word by word， and the obtained attention weight and the context representation were fused as the input of the mLSTM， so that the context representation with higher correlation with the aspect word was obtained. Then， the grammatical distance was introduced to obtain a context which was more grammatically related to the aspect word， so as to obtain more contextual features to guide the modeling of the aspect word， and obtain the aspect representation through the aspect masking layer. Finally， in order to exchange information， location weights， context representations and aspect representations were combined， thereby obtaining the features for sentiment analysis. Experimental results on Twitter， REST14 and LAP14 datasets show that compared with Aspect-Specific Graph Convolutional Network （ASGCN）， mLSTM-GCN has the accuracy improved by 1.32， 2.50 and 1.63 percentage points， respectively， and has the Macro-F1 score improved by 2.52， 2.19 and 1.64 percentage points， respectively. Therefore， mLSTM-GCN can effectively reduce the probability of mismatch between aspect words and irrelevant context， and improve the classification effect.

Review of fine-grained image categorization

SHEN Zhijun, MU Lina, GAO Jing, SHI Yuanhang, LIU Zhiqiang

2023, 43(1):  51-60.  DOI: 10.11772/j.issn.1001-9081.2021122090

Asbtract ( )   HTML ( )   PDF (2674KB) ( )  

References | Related Articles | Metrics

The fine-grained image has characteristics of large intra-class variance and small inter-class variance， which makes Fine-Grained Image Categorization （FGIC） much more difficult than traditional image classification tasks. The application scenarios， task difficulties， algorithm development history and related common datasets of FGIC were described， and an overview of related algorithms was mainly presented. Classification methods based on local detection usually use operations of connection， summation and pooling， and the model training was complex and had many limitations in practical applications. Classification methods based on linear features simulated two neural pathways of human vision for recognition and localization respectively， and the classification effect is relatively better. Classification methods based on attention mechanism simulated the mechanism of human observation of external things， scanning the panorama first， and then locking the key attention area and forming the attention focus， and the classification effect was further improved. For the shortcomings of the current research， the next research directions of FGIC were proposed.

Dense crowd detection algorithm based on Faster R-CNN

ZOU Bin, ZHANG Cong

2023, 43(1):  61-66.  DOI: 10.11772/j.issn.1001-9081.2021111950

Asbtract ( )   HTML ( )   PDF (3411KB) ( )  

References | Related Articles | Metrics

In order to improve the accuracy of crowd detection in crowded scenes， a dense crowd detection algorithm based on improved Faster Region-based Convolutional Neural Network （Faster R-CNN） was proposed. Firstly， the spatial and channel attention mechanisms were added to feature extraction stage and Strong-Bidirectional Feature Pyramid Network（S-BiFPN） was used to replace the multi-scale Feature Pyramid Network （FPN） in the original network， so that the network was able to autonomously learn important features and the extraction of deep image features was strengthened. Secondly， Multi-Instance Prediction （MIP） algorithm was introduced to predict instances， thus avoiding the model’s missed detection of targets in crowded scenes. Finally， Non-Maximum Suppression （NMS） in the model was optimized， and an additional Intersection over Union （IoU） threshold was added to accurately suppress the interference items of the detection results. Experimental results on the open source dense crowd detection dataset show that compared with the original Faster R-CNN algorithm， the proposed algorithm has the Average Precision （AP） increased by 5.6%， and Jaccard index value increased by 3.2%. The proposed algorithm has high detection precision and stability， which can meet the needs of crowd detection in dense scenes.

License plate detection algorithm in unrestricted scenes based on adaptive confidence threshold

LIU Xiaoyu, CHEN Huaixin, LIU Biyuan, LIN Ying, MA Teng

2023, 43(1):  67-73.  DOI: 10.11772/j.issn.1001-9081.2021111974

Asbtract ( )   HTML ( )   PDF (2162KB) ( )  

References | Related Articles | Metrics

Aiming at the problem of low generalization of the license plate detection model， which makes it difficult to reuse in different application scenes of smart transportation， a license plate detection algorithm in unrestricted scenes based on adaptive confidence threshold was proposed. Firstly， a multi-prediction head network model was constructed， in it， the segmentation prediction head was used to reduce the model reuse pre-processing work， the adaptive confidence threshold prediction head was used to improve the model detection ability， and the multi-scale fusion mechanism and bounding box regression prediction head were used to improve the model generalization ability. Secondly， a differentiable binary network training method was adopted to learn model parameters through differentiable binary transformation combined with the training of classification confidence and confidence threshold. Finally， the Connectivity Aware Non-Maximum Suppression （CANMS） method was used to improve the post-processing speed of license plate detection， and the lightweight network ResNet18 was introduced as the backbone network of feature extraction to reduce the model parameters and further improve the detection speed. Experimental results show that in 6 scenes with different constraints in Chinese City Parking Dataset （CCPD）， the proposed algorithm can achieve the average precision of 99.5% and the recall of 99.8%， and achieves the efficient detection rate of 70 frames per second， which are better than the performance of anchor-based algorithms such as Faster Region-Conventional Neural Network （Faster R-CNN） and Single Shot MultiBox Detector （SSD）. On the three supplementary scene test sets， the license plate detection accuracy of the proposed algorithm is higher than 90% in unrestricted scenes with different resolutions， different shooting distances， and different shooting angles of pitch. Therefore， the proposed algorithm has good detection performance and generalization ability in unrestricted scenes， and can meet the requirements of model reuse.

Real‑time detection method of traffic information based on lightweight YOLOv4

GUO Keyou, LI Xue, YANG Min

2023, 43(1):  74-80.  DOI: 10.11772/j.issn.1001-9081.2021101849

Asbtract ( )   HTML ( )   PDF (3243KB) ( )  

References | Related Articles | Metrics

Aiming at the problem of vehicle objection detection in daily road scenes， a real?time detection method of traffic information based on lightweight YOLOv4 （You Only Look Once version 4） was proposed. Firstly， a multi?scene and multi?period vehicle object dataset was constructed， which was preprocessed by K?means++ algorithm. Secondly， a lightweight YOLOv4 detection model was proposed， in which the backbone network was replaced by MobileNet?v3 to reduce the number of parameters of the model， and the depth separable convolution was introduced to replace the standard convolution in the original network. Finally， combined with label smoothing and annealing cosine algorithms， the activation function Leaky Rectified Linear Unit （LeakyReLU） was used to replace the original activation function in the shallow network of MobileNet?v3 in order to optimize the convergence effect of the model. Experimental results show that the lightweight YOLOv4 has the weight file of 56.4 MB， the detection rate of 85.6 FPS （Frames Per Second）， and the detection precision of 93.35%， verifying that the proposed method can provide the reference for the real?time traffic information detection and its applications in real road scenes.

Improved YOLOv3 target detection based on boundary limit point features

LI Kewen, YANG Jiantao, HUANG Zongchao

2023, 43(1):  81-87.  DOI: 10.11772/j.issn.1001-9081.2021111999

Asbtract ( )   HTML ( )   PDF (2069KB) ( )  

References | Related Articles | Metrics

The problems of large number of targets， small scale and high-overlapping lead to low accuracy and difficulty in target detection. In order to improve the precision of target detection and avoid missed detection and false detection as much as possible， an improved YOLOv3 target detection algorithm based on boundary limit point features was proposed. Firstly， a boundary enhancement operator Border was introduced to adaptively extract boundary features from the limit points of the boundary to enhance the features of the existing points and improve the accuracy of target positioning. Then， the precision of target detection was further improved by increasing the target detection scale， refining the feature map， and enhancing the fusion of the feature image deep and shallow semantic information. Finally， based on the target instance characteristics in target detection and the improved network model， the Complete Intersection over Union （CIoU） function was introduced to improve the original YOLOv3 loss function， thereby improving the convergence speed and recall of the detection box. Experimental results show that compared with the original YOLOv3 target detection algorithm， the improved YOLOv3 target detection algorithm has the Average Precision increased by 3.9 percentage points ， and has the detection speed similar to the original algorithm， verifying that it can effectively improve the target detection ability of models.

Arthropod object detection method based on improved Faster RCNN

GUO Zihao, DONG Lele, QU Zhijian

2023, 43(1):  88-97.  DOI: 10.11772/j.issn.1001-9081.2021101838

Asbtract ( )   HTML ( )   PDF (4771KB) ( )  

References | Related Articles | Metrics

Arthropod object detection in natural environment has characteristics of complex object background， large scale difference， and dense objects，resulting in poor object detection accuracy and precision. Therefore， an arthropod object detection method was proposed based on the improved Faster RCNN model， namely AROD RCNN （ARthropod Object Detection RCNN）. Firstly， a Supervised Parallel mechanism with Spatial and Channel ATtention modules （SPSCAT） was designed to improve the accuracy of arthropod object detection in the environment with complex background. Then， the second-generation deformable convolution was introduced to reconstruct the convolutional layer with C1~C5 blocks in ResNet50， and the Feature Pyramid Network （FPN） was adopted to perform feature fusion on the C2~C6 blocks in ResNet50 to solve the problem that large difference in object scale affected detection accuracy. Finally， the Dense Local Regression （DLR） method was used to improve the regression stage， thereby improving the accuracy of the model regression. Experimental results show that on ArTaxOr （Arthropod Taxonomy Orders Object Detection） dataset， the proposed method has the mean Average Precision （mAP） of 0.717， which is 0.453 higher than that of the original model， and has the recall reached 0.787. It can be seen that the proposed method can effectively solve the problems of object occlusion and complex background， and performs well in the detection of dense arthropod objects and small arthropod objects.

Target detection of Ochotona curzoniae based on embedded Jetson TX2

CHEN Haiyan, JIA Mingming, ZHAO Wenli, WANG Chanfei

2023, 43(1):  98-103.  DOI: 10.11772/j.issn.1001-9081.2021101857

Asbtract ( )   HTML ( )   PDF (2996KB) ( )  

References | Related Articles | Metrics

Target detection of Ochotona curzoniae is the basis for its population statistics and population dynamic changes research， but the traditional intelligent monitoring systems of Ochotona curzoniae has a large target detection hardware equipment and weak mobility in sampling and collecting data. To solve the above problems， an improved YOLOv3? based target detection method that can be deployed to the portable device Jetson TX2 was proposed. The lightweight Ochotona curzoniae target detection model was constructed by replacing Darknet53， backbone network of YOLOv3， with MobileNet and using pruning and fine?tuning methods to further reduce the model size. Next， the model was deployed on the portable target detection device Jetson TX2. Experimental results of Ochotona curzoniae target detection in natural scenes show that the proposed method has the detection Average Precision （AP）， detection Frames per Second （FPS） and model size of 97.36%， 36 and 14.88 MB， respectively， which are better than those of the improved YOLOv3 model without pruning and the original YOLOv3 model； with the AP only reduced by 1.05 percentage points， the proposed method has the detection speed improved by 620% and the model size compressed by 93.67%， which can be deployed to portable devices for real-time and accurate detection of Ochotona curzoniae.

Data science and technology

Local community detection algorithm based on Monte-Carlo iterative solving strategy

LI Zhanli, LI Ying, LUO Xiangyu, LUO Yingxiao

2023, 43(1):  104-110.  DOI: 10.11772/j.issn.1001-9081.2021111942

Asbtract ( )   HTML ( )   PDF (1690KB) ( )  

References | Related Articles | Metrics

Aiming at the problems of premature convergence and low recall caused by using greedy strategy for community expansion in the existing local community detection algorithms， a local community detection algorithm based on Monte-Carlo iterative solving strategy was proposed. Firstly， in the community expansion stage of each iteration， the selection probabilities were given to all adjacent candidate nodes according to the contribution ratio of each node to the community tightness gain， and one node was randomly selected to join the community according to these probabilities. Then， in order to avoid random selection causing the expansion direction to deviate from the target community， it was determined whether the node elimination mechanism was triggered in this round of iteration according to the changes in community quality. If it was triggered， the similarity sum of each node joining the community and other nodes in the community was calculated， the elimination probabilities were assigned according to the reciprocal of the similarity sum， a node was randomly eliminated according to these probabilities. Finally， whether to continue the iteration was judged on the basis of whether the community size increased in a given number of recent iteration rounds. Experimental results show that， on three real network datasets， compared to Local Tightness Expansion （LTE） algorithm， Clauset algorithm， Common Neighbors with Weighted Neighbor Nodes （CNWNN） algorithm and Fuzzy Similarity Relation （FSR） algorithm， the proposed algorithm has the F-score value of local community detection results increased by 32.75 percentage points， 17.31 percentage points， 20.66 percentage points and 25.51 percentage points respectively， and can effectively avoid the influence of the location of the query node in the community on the local community detection results.

Range query algorithm for large scale moving objects in distributed environment

MA Yongqiang, CHEN Xiaomeng, YU Ziqiang

2023, 43(1):  111-121.  DOI: 10.11772/j.issn.1001-9081.2021101853

Asbtract ( )   HTML ( )   PDF (3320KB) ( )  

References | Related Articles | Metrics

Continuous range queries over moving objects is essential to many location-based services. Aiming at this issue， a distributed search method was proposed for processing concurrent range queries over large scale moving objects. Firstly， formed by a Global Grid Index （GGI） and a local elastic quadtree， a Distributed Dynamic Index （DDI） structure was proposed. Then， a Distributed Search Algorithm （DSA） was proposed based on DDI structure. At the first time， an incremental strategy of updating the query results as objects and query points continuously changed their locations was introduced by DSA. After that， in the incremental update process， a shared computing optimization strategy for multiple concurrent queries was introduced， to incrementally search the range query results of the moving object according to the existing calculation results. Finally， three moving object datasets with different spatial distributions were simulated on the basis of the German road network， and NS （Naive Search）， GI （Grid Index） and Distributed Hybrid Index （DHI） were compared with the proposed algorithm. The results show that compared with DHI， the comparison algorithm with the best performance， DSA decreases the initial query time by 22.7%， while drops the incremental query time by 15.2%， verifying that DSA is superior to the comparison algorithms.

Cyber security

Improved practical Byzantine fault tolerance consensus algorithm based on Raft algorithm

WANG Jindong, LI Qiang

2023, 43(1):  122-129.  DOI: 10.11772/j.issn.1001-9081.2021111996

Asbtract ( )   HTML ( )   PDF (2834KB) ( )  

References | Related Articles | Metrics

Since Practical Byzantine Fault Tolerance （PBFT） consensus algorithm applied to consortium blockchain has the problems of insufficient scalability and high communication overhead， an improved practical Byzantine fault tolerance consensus algorithm based on Raft algorithm named K-RPBFT （K-medoids Raft based Practical Byzantine Fault Tolerance） was proposed. Firstly， blockchain was sharded based on K-medoids clustering algorithm， all nodes were divided into multiple node clusters and each node cluster constituted to a single shard， so that global consensus was improved to hierarchical multi-center consensus. Secondly， the consus between the cluster central nodes of each shard was performed by adopting PBFT algorithm， and the improved Raft algorithm based on supervision nodes was used for intra-shard consensus. The supervision mechanism in each shard gave a certain ability of Byzantine fault tolerance to Raft algorithm and improved the security of the algorithm. Experimental analysis shows that compared with PBFT algorithm， K-RPBFT algorithm greatly reduces the communication overhead and consensus latency， improves the consensus efficiency and throughput while having Byzantine fault tolerance ability， and has good scalability and dynamics， so that the consortium blockchain can be applied to a wider range of fields.

Fair and verifiable multi-keyword ranked search over encrypted data based on blockchain

PANG Xiaoqiong, WANG Yunting, CHEN Wenjun, JIANG Pan, GAO Yanan

2023, 43(1):  130-139.  DOI: 10.11772/j.issn.1001-9081.2021111904

Asbtract ( )   HTML ( )   PDF (1334KB) ( )  

References | Related Articles | Metrics

In view of the high cost as well as the limitation of retrieval function of the existing searchable encryption schemes based on blockchain to realize result verification and fair payment， a multi-keyword ranked search scheme supporting verification and fair payment was proposed based on blockchain. In the proposed scheme， the Cloud Service Provider （CSP） was used to store the encrypted index tree and perform search operations， and a lookup table including verification certificates was constructed to assist the smart contract to complete the verification of retrieval results and fair payment， which reduced the complexity of smart contract execution and saved time as well as expensive cost. In addition， the index of balanced binary tree structure was constructed by combining vector space model and Term Frequency-Inverse Document Frequency （TF-IDF）， and the index and query vectors were encrypted by using secure K-nearest neighbor， which realized the multi-keyword ranked search supporting dynamic update. Security and performance analysis show that the proposed scheme is secure and feasible in the blockchain environment and under the known ciphertext model. Simulation results show that the proposed scheme can achieve result verification and fair payment with acceptable cost.

Secure consensus of multi‑agent systems based on event‑triggered impulsive control

GAO An’an, HU Aihua, JIANG Zhengxian

2023, 43(1):  140-146.  DOI: 10.11772/j.issn.1001-9081.2021122037

Asbtract ( )   HTML ( )   PDF (2452KB) ( )  

References | Related Articles | Metrics

The mean square bounded consensus problem of multi?agent systems under deception attacks was studied， and an event?triggered impulsive control method with upper bound of trigger time was designed based on the fixed?time impulsive control and event?triggered mechanism. By utilizing the Lyapunov stability theory， graph theory， and linear matrix inequality techniques， the sufficient conditions for the mean square bounded consensus of the multi?agent systems were obtained. Moreover， it was verified that the proposed event?triggered impulsive control method can automatically adjust the impulsive time intervals， and realize the secure consensus quickly. Finally， the numerical simulation results further verified the validity of the theoretical results.

Analysis and improvement of certificateless signature scheme

ZHAO Hong, YU Shuhan, HAN Yanyan, LI Zhaobin

2023, 43(1):  147-153.  DOI: 10.11772/j.issn.1001-9081.2021111919

Asbtract ( )   HTML ( )   PDF (910KB) ( )  

References | Related Articles | Metrics

For nine certificateless signature schemes proposed by Y L Tang， et al. （TANG Y L， WANG F F， YE Q， et al. Improved provably secure certificateless signature scheme. Journal of Beijing University of Posts and Telecommunications， 2016， 39（1）： 112-116）， firstly， the linearized equation analysis method was used. It was found that there was a linear relationship between the public keys in all schemes. This defect was exploited to complete a signature forgery attack on all schemes. Secondly， in order to break the linear relationship between the public keys， the method of modifying the parameters of hash function was used to improve the scheme， and the security of the improved scheme was proved under the random oracle model. Thirdly， a public key construction format of certificateless signature scheme was proposed. The signature scheme constructed by this format could not be attacked by adversaries using public key replacement. Finally， the efficiency of the improved scheme was compared with those of the existing certificateless signature schemes through simulation. Experimental results show that the improved scheme promotes the security without reducing the computational efficiency.

Lattice-based hierarchical certificateless proxy signature scheme

NONG Qiang, ZHANG Bangbang, OUYANG Yuhao

2023, 43(1):  154-159.  DOI: 10.11772/j.issn.1001-9081.2021111945

Asbtract ( )   HTML ( )   PDF (749KB) ( )  

References | Related Articles | Metrics

Existing certificateless proxy signature schemes based on classical number theory problem assumptions cannot resist to quantum computer attacks， and when these schemes are applied to systems with a large number of users， there are limitations such as single point of failure and low scalability. Aiming at these problems， a lattice-based hierarchical certificateless proxy signature scheme was proposed. Firstly， the rejection sampling technology and trapdoor-free technology were used to improve the computational efficiency of key generation. Secondly， the mutual authentication was performed by the original signers and proxy signers at different levels by exchanging randomly selected matrices， and then the proxy authorization was realized. Finally， the security of this scheme was proved under the of the Small Integer Solution （SIS） hard problem assumption in the random oracle model. Compared with the existing proxy signature schemes， the proposed scheme allows signers coming from different levels and belonging to different Key Generation Centers （KGCs）. The performance evaluation experimental results show that in the proposed scheme， the public key size is a constant， the overhead of proxy signature and verification is independent of the level， and the proxy key size and the signature size are not hierarchical linear quantities， so that this scheme can better meet the needs of large-scale distributed heterogeneous networks for load balancing， and is efficient and feasible.

Intrusion detection method for wireless sensor network based on bidirectional circulation generative adversarial network

LIU Yongmin, YANG Yujin, LUO Haoyi, HUANG Hao, XIE Tieqiang

2023, 43(1):  160-168.  DOI: 10.11772/j.issn.1001-9081.2021112001

Asbtract ( )   HTML ( )   PDF (2098KB) ( )  

References | Related Articles | Metrics

Aiming at the problems of low detection accuracy and poor generalization ability of Wreless Sensor Network （WSN） intrusion detection methods on imbalanced datasets with discrete high-dimensional features， an intrusion detection method for WSN based on Bidirectional Circulation Generative Adversarial Network was proposed， namely BiCirGAN. Firstly， Adversarially Learned Anomaly Detection （ALAD） was introduced to improve the understandability of the original features by reasonably representing the high-dimensional， discrete original features through the latent space. Secondly， the bidirectional circulation adversarial structure was adopted to ensure the consistency of bidirectional circulation in real space and latent space， thereby ensuring the stability of Generative Adversarial Network （GAN） training and improving performance of anomaly detection. At the same time， Wasserstein distance and spectral normalization optimization methods were introduced to improve the objective function of GAN to further solve the problems of mode collapse of GAN and lack of diversity of generators. Finally， because the statistical properties of intrusion attack data changed in an unpredictable way over time， a full connection layer network with Dropout operation was established to optimize the anomaly detection results. Experimental results on KDD99， UNSW-NB15 and WSN_DS datasets show that compared to Anomaly detection with GAN （AnoGAN）， Bidirectional GAN （BiGAN）， Multivariate Anomaly Detection with GAN （MAD-GAN） and ALAD methods， BiCirGAN has a 3.9% to 33.0% improvement in detection accuracy， and the average inference speed is 4.67 times faster than that of ALAD method.

Semi-generative video steganography scheme based on deep convolutional generative adversarial net

LIN Yangping, LIU Jia, CHEN Pei, ZHANG Mingshu, YANG Xiaoyuan

2023, 43(1):  169-175.  DOI: 10.11772/j.issn.1001-9081.2021112035

Asbtract ( )   HTML ( )   PDF (3023KB) ( )  

References | Related Articles | Metrics

Generative steganography hides secret messages by generating sufficiently natural or true samples with secret，which is a hot research topic in information hiding， but there is little research in the field of video steganography. Combined with the idea of digital Cardan grille， a semi-generative video steganography scheme based on Deep Convolutional Generative Adversarial Net （DCGAN） was proposed. In this scheme， a dual-stream video generation network based on DCGAN was designed to generate three parts of videos： dynamic foreground， static background and spatio-temporal mask， and different videos were produced by the generation network driven by random noise. The sender in this scheme was able to set the steganography threshold and adaptively generate a digital Cardan grille in the mask， then the obtain digital cardan grille was used as the key for steganography and extraction； at same time， with the foreground as the carrier， the optimal embedding of information was realized. Experimental results show that the video-with-secret generated by the proposed scheme has good visual quality， with a Frechet Inception Distance score （FID） of 90， and the embedding capacity of the scheme is better than those of the existing generative steganography schemes， up to 0.11 bpp. It can be seen that the proposed scheme can transmit secret messages more efficiently.

Advanced computing

Adaptive scheduling strategy based on deadline under cloud platform

WU Renbiao, ZHANG Zhenchi, JIA Yunfei, QIAO Han

2023, 43(1):  176-184.  DOI: 10.11772/j.issn.1001-9081.2021112018

Asbtract ( )   HTML ( )   PDF (2505KB) ( )  

References | Related Articles | Metrics

Aiming at the problem that the response speed and the completion time of the task cannot be taken into account at the same time when scheduling tasks in a shared cluster， an adaptive scheduling algorithm based on deadline was proposed. In the algorithm， based on the deadline submitted by the user， the appropriate computing resources were allocated adaptively according to the execution progress of the tasks. Different from that fixed resource parameters were submitted by users in the traditional scheduling methods， in this algorithm， tasks with high priority would be executed with preemptive scheduling under resource constraints. Preemptive scheduling was used to ensure the Quality of Service （QoS）， and additional resources would be allocated to compensate the preempted tasks after the preemption process. The task scheduling experimental results on the Spark platform show that compared with the scheduling algorithm under Yet Another Resource Negotiator （YARN） framework， the proposed algorithm can control the response speed of short tasks strictly and shorten the task completion time of long jobs by 35%.

Optimal task scheduling method based on satisfiability modulo theory for multiple processors with communication delay

JIANG Songyan, LIAO Xiaojuan, CHEN Guangzhu

2023, 43(1):  185-191.  DOI: 10.11772/j.issn.1001-9081.2021111862

Asbtract ( )   HTML ( )   PDF (1508KB) ( )  

References | Related Articles | Metrics

It is of great significance in the scheduling theory and practice of parallel computing to achieve the minimum execution time of task graphs with communication delays on homogeneous multiple processors. For the task graph scheduling problem with communication delay， an optimization method based on Satisfiability Modulo Theory （SMT） was proposed. Firstly， constraints such as processor mapping constraints and task execution order were encoded， thus the task graph scheduling problem was transformed into an SMT problem. Then， the SMT solver was called to search the feasible solution space to determine the optimal solution of the problem. In the constraint encoding phase， integer variables were introduced to represent the mapping relationships between tasks and processors， thereby reducing the complexity of processor constraint encoding. In the solver calling phase， the constraints of independent tasks were added to reduce the search space of the solver and further improve the search efficiency of the optimal solution. Experimental results show that compared with the original SMT method， the improved SMT method reduces the average solving time by 65.9% and 53.8% in timeout experiments of 20 seconds and 1 minute， respectively， and achieves a greater efficiency advantage when the number of processors is relatively large. The improved SMT method can effectively solve the task graph scheduling problem with communication delay， especially be suitable for scheduling scenarios with a large number of processors.

Bald eagle search optimization algorithm with golden sine algorithm and crisscross strategy

ZHAO Peiwen, ZHANG Damin, ZHANG Linna, ZOU Chengcheng

2023, 43(1):  192-201.  DOI: 10.11772/j.issn.1001-9081.2021111868

Asbtract ( )   HTML ( )   PDF (1555KB) ( )  

References | Related Articles | Metrics

Aiming at the disadvantages of traditional Bald Eagle Search optimization algorithm （BES）， such as easy to fall into the local optimum and slow convergence， a BES with Golden Sine Algorithm （Gold-SA） and crisscross strategy （GSCBES） was proposed. Firstly， the position update formula based on inertia weight was set in the traditional BES search stage. Then， Gold-SA was introduced in the stage of predation. Finally， the crisscross strategy was introduced to modify the global optimum and population. The optimization ability of the proposed algorithm was evaluated by the simulation experiments on 11 Benchmark functions， CEC2014 functions and by using Wilcoxon rank sum test. The results show that the proposed algorithm converges faster. At the same time， the weights and thresholds of Back Propagation （BP） neural network were assigned by the proposed algorithm， and the optimized BP neural network model was used in the prediction of air quality， the values of Mean Absolute Error （MAE）， Root Mean Square Error （RMSE）， Mean Square Error （MSE）， and Mean Absolute Percentage Error （MAPE） are smaller than those of BP neural network model and Particle Swarm Optimization （PSO） based BP neural network model，and the prediction accuracy is improved.

Improved QMIX algorithm from communication and exploration for multi-agent reinforcement learning

DENG Huiyi, LI Yongzhen, YIN Qiyue

2023, 43(1):  202-208.  DOI: 10.11772/j.issn.1001-9081.2021111886

Asbtract ( )   HTML ( )   PDF (1867KB) ( )  

References | Related Articles | Metrics

Non-stationarity that breaks the Markov assumption followed by most single-agent reinforcement learning algorithms is one of the main challenges in multi-agent environment， making each agent may be caught in an infinite loop caused by the environment created by the other agents during the learning process. To solve above problem， the implementation method of Centralized Training with Decentralized Execution （CTDE） structure in reinforcement learning was studied， and from two perspectives of agent communication and exploration， the QMIX algorithm was improved by introducing a Variance Control-Based （VBC） communication model and a curiosity mechanism. The proposed algorithm was validated in micro control scenarios of StarCraft Ⅱ Learning Environment （SC2LE）. Experimental results show that the proposed algorithm can improve the performance and obtain a training model with higher convergence speed compared to QMIX algorithm.

Network and communications

Sink location algorithm of power domain nonorthogonal multiple access for real-time industrial internet of things

SUN Yuan, SHEN Wenjian, NI Pengbo, MAO Min, XIE Yaqi, XU Chaonong

2023, 43(1):  209-214.  DOI: 10.11772/j.issn.1001-9081.2021111946

Asbtract ( )   HTML ( )   PDF (2234KB) ( )  

References | Related Articles | Metrics

Aiming at the shortcoming of large access delay in industrial Internet of Things （IoT）， a sink location algorithm of Power Domain NonOrthogonal Multiple Access （PD-NOMA） for real-time industrial IoT was proposed. In this algorithm， based on the PD-NOMA technology， the location of the sink was used as an optimization method to minimize access delay by realizing power division multiplexing among users as much as possible. Firstly， for any two users， an assertion that the decodable area of the qualified sink must be a circle if parallel transmissions are successful was proven， and therefore， the decodable area set of the sink was able to be obtained by combining all of the combinations of two users， and every minimal intersection of the area set must be a convex region. So， the optimal location of the sink must be included in these minimal intersection areas. Secondly， for each minimal intersection area where the sink was deployed， the minimum number of chain partition of the network generation graph in the area was computed and used as the metric for evaluating the access delay. Finally， the optimal location of the sink was determined by comparing these minimum number of chain partitioning. Experimental results show that when the decoding threshold is 2 and the number of users is 30， the average access delay of the proposed algorithm is about 36.7% of that of the classic time division multiple access， and besides， it can be decreased almost linearly with the decrease of the decoding threshold and the increase of the channel decay factor. The proposed algorithm can provide reference from the access layer perspective for massive ultra-reliable low-latency communications.

Computer software technology

Structural testing methods of disjunctive answer set programs

YANG Dong, WANG Yisong

2023, 43(1):  215-220.  DOI: 10.11772/j.issn.1001-9081.2021111891

Asbtract ( )   HTML ( )   PDF (712KB) ( )  

References | Related Articles | Metrics

Focused on the lack of basic theories for structural testing of disjunctive answer set programs， concepts of structured test coverage for disjunctive answer set programs was proposed systematically. Firstly， the test cases of disjunctive answer set programs were defined， and the logic rules in the program were determined to be the main test entities of disjunctive answer set programs. Then， the basic concepts such as rule coverage， definition coverage and loop coverage were constructed for different test targets such as rule header， rule body， and rule set to simulate the concepts such as statement coverage and branch coverage in structural testing. Finally， the calculation formula of test coverage rate for disjunctive answer set programs was proposed， and different coverage calculation methods under different types of coverage were illustrated in samples. At the same time， some special characteristics and key indicators of the disjunctive answer set programs were discussed.

Test case prioritization approach based on historical data and multi-objective optimization

LI Xingjia, YANG Qiuhui, HONG Mei, PAN Chunxia, LIU Ruihang

2023, 43(1):  221-226.  DOI: 10.11772/j.issn.1001-9081.2021112015

Asbtract ( )   HTML ( )   PDF (1305KB) ( )  

References | Related Articles | Metrics

To improve the error detection efficiency and the benefit of regression testing of test case sequence， a test case prioritization approach based on historical data and multi-objective optimization was proposed. Firstly， the test case set was clustered according to the text topic similarity and code coverage similarity of test cases， and the association rules were mined for execution failure relationships between test cases according to the historical execution information， thereby preparing for the subsequent process. Then， the multi-objective optimization algorithm was used to sort the test cases in each cluster. After that， the final sorting sequence was generated to separate the similar test cases. Finally， the association rules between test cases were used to dynamically adjust the execution order of test cases， so that the test cases that may fail were executed with priority， so as to further improve the efficiency of defect detection. Compared with random search approach， the approach based on clustering， the approach based on topic model， the approach based on association rules and multi-objective optimization， the proposed approach has the average value of Average Percentage of Faults Detected （APFD） increased by 12.59%， 5.98%， 3.01% and 2.95%， respectively， and has the average value of APFD cost-cognizant （APFDc） increased by 17.17%， 5.04%， 5.08% and 8.21%， respectively. Experimental results show that the proposed approach can improve the benefit of regression testing effectively.

Multimedia computing and computer simulation

Semi‑supervised end‑to‑end fake speech detection method based on time‑domain waveforms

FANG Xin, HUANG Zexin, ZHANG Yuhan, GAO Tian, PAN Jia, FU Zhonghua, GAO Jianqing, LIU Junhua, ZOU Liang

2023, 43(1):  227-231.  DOI: 10.11772/j.issn.1001-9081.2021101845

Asbtract ( )   HTML ( )   PDF (6257KB) ( )  

References | Related Articles | Metrics

The fake speech produced by modern speech synthesis and timbre conversion systems poses a serious threat to the automatic speaker recognition system. Most of the existing fake speech detection systems perform well for the known attack types in the training process， but degrades significantly in detecting unknown attack types in practical applications. Therefore， combined with the recently proposed Dual?Path Res2Net （DP?Res2Net）， a semi?supervised end?to?end fake speech detection method based on time?domain waveforms was proposed. Firstly， semi?supervised learning was adopted for domain transfer to reduce the difference of data distribution between training set and test set. Then， for feature engineering， time-domain sampling points were input into DP?Res2Net directly， which increased the local multi?scale information and made full use of the dependence between audio segments. Finally， the embedded tensors were obtained to judge fake speech from natural speech after the input features going through the shallow convolution module， feature fusion module and global average pooling module. The performance of the proposed method was evaluated on the publicly available ASVspoof 2021 Speech Deep Fake evaluation set as well as the dataset VCC （Voice Conversion Challenge）. Experimental results show that the Equal Error Rate （EER） of the proposed method is 19.97%， which is 10.8% less than that of the official optimal baseline system， verifying that the semi?supervised end?to?end fake speech detection method based on time?domain waveforms is effective when recognizing unknown attacks and has higher generalization capability.

Section steel surface defect detection algorithm based on cascade neural network

YU Haitao, LI Jiansheng, LIU Yajiao, LI Fulong, WANG Jiang, ZHANG Chunhui, YU Lifeng

2023, 43(1):  232-241.  DOI: 10.11772/j.issn.1001-9081.2021111940

Asbtract ( )   HTML ( )   PDF (4174KB) ( )  

References | Related Articles | Metrics

Deep learning has superior performance in defect detection， however， due to the low defect probability， the detection process of defect-free images occupies most of the calculation time， which seriously limits the overall effective detection speed. In order to solve the above problem， a section steel surface defect detection algorithm based on cascade network named SDNet （Select and Detect Network） was proposed. The proposed algorithm was divided into two stages： the pre-inspection stage and the precise detection stage. In the pre-inspection stage， the lightweight ResNet pre-inspection network based on Depthwise Separable Convolution （DSC） and multi-scale parallel convolution was used to determine whether there were defects in the surface image of the section steel. In the precise detection stage， the YOLOv3 was used as the baseline network to accurately classify and locate the defects in the image. In addition， the improved Atrous Spatial Pyramid Pooling （ASPP） module and dual attention module were introduced in the backbone feature extraction network and prediction branches to improve the network detection performance. Experimental results show that the detection speed and the accuracy of SDNet on 1 024 pixel×1 024 pixel images reach 120.63 frames per second and 92.1% respectively. Compared to the original YOLOv3 algorithm， the proposed algorithm has the detection speed of about 3.7 times and the detection precision improved by 10.4 percentage points. The proposed algorithm can be applied to the rapid detection of section steel surface defects.

Strip steel surface defect detection by YOLOv5 algorithm fusing frequency domain attention mechanism and decoupled head

SUN Zeqiang, CHEN Bingcai, CUI Xiaobo, WANG Lei, LU Yanuo

2023, 43(1):  242-249.  DOI: 10.11772/j.issn.1001-9081.2021111926

Asbtract ( )   HTML ( )   PDF (3035KB) ( )  

References | Related Articles | Metrics

Aiming at the low detection precision of strip steel surface defects in actual scenarios， which is prone to missed detection and false detection， a YOLOv5-CFD model consisted of CSPDarknet53， Frequency channel attention Network （FcaNet） and Decoupled head was constructed to detect strip steel defects more accurately. Firstly， Fuzzy C-Means （FCM） algorithm was used to cluster anchor boxes in NEU-DET hot-rolling strip steel surface defect detection dataset published by Northeastern University to optimize the matching degree between the prior box and the ground-truth box. Secondly， in order to extract the rich detailed information of the target area， the frequency domain channel attention module FcaNet （Frequency channel attention Network） was added to the original YOLOv5 algorithm. Finally， the decoupled head was used to separate the classification and regression tasks. Experimental results on NEU-DET dataset show that with introducing a small number of parameters to the original YOLOv5 algorithm， the improved YOLOv5 algorithm has the detection precision increased by 4.2 percentage points， the detection mean Average Precision （mAP） of 85.5%； and the detection speed reaches 27.71 Frames Per Second （FPS）， which is not much different from the original YOLOv5 so that YOLOv5-CFD can meet the real-time detection requirements.

Print defect detection method based on deep comparison network

WANG Youxin, CHEN Bin

2023, 43(1):  250-258.  DOI: 10.11772/j.issn.1001-9081.2021111920

Asbtract ( )   HTML ( )   PDF (2660KB) ( )   PDF（mobile） (3120KB) ( 8 )  

References | Related Articles | Metrics

The print defect detection methods based on traditional image processing technology have poor robustness and the object detection methods based on deep learning are not completely suitable for the detection tasks of print defects. In order to solve the problems above， the comparison ideas in template matching method were combined with the semantic features in deep learning， and a Deep Comparison Network （CoNet） used for the detection tasks of print defects was proposed. Firstly， the Deep Comparison Module （DCM） adopting Siamese structure was proposed to mine the semantic relationship between the detection image and the reference image through extracting and fusing the feature maps of them in the semantic space. Then， based on the feature pyramid structure with asymmetric dual channels， the Multi-scale Change Detection Module （MsCDM） was proposed to locate and classify print defects. On the public printed circuit board defect dataset DeepPCB and dataset of Lijin defects， the average values of mean Average Precision （mAP） of CoNet are 99.1% and 69.8% respectively， compared with the two baseline models Max-Pooling Group Pyramid Pooling （MP-GPP） and Change-Detection Single Shot Detector （CD-SSD）， which are increased by 0.4， 3.5 percentage points and 0.7， 2.4 percentage points respectively， and the detection accuracy of CoNet is higher. Besides， when the resolution of input image is 640×640， the average time consumption of CoNet is 35.7 ms， showing that it can absolutely meet the real-time requirements of industrial detection tasks.

Bilinear involution neural network for image classification of fundus diseases

YANG Honggang, CHEN Jiejie, XU Mengfei

2023, 43(1):  259-264.  DOI: 10.11772/j.issn.1001-9081.2021111932

Asbtract ( )   HTML ( )   PDF (2180KB) ( )  

References | Related Articles | Metrics

Due to the high complexity， weak individual differences， and short inter-class distances of fundus image features， pure Convolutional Neural Networks （CNNs） and attention based networks cannot achieve satisfactory accuracy in fundus disease image classification tasks. To this end， Attention Bilinear Involution Neural Network （ABINN） model was implemented for fundus disease image classification by using the involution operator. The parameter amount of ABINN model was only 11% of that of the traditional Bilinear Convolutional Neural Network （BCNN） model. In ABINN model， the underlying semantic information and spatial structure information of the fundus image were extracted and the second-order features of them were fused. It is an effective parallel connection between CNN and attention method. In addition， two instantiation methods for attention calculation based on involution operator， Attention Subnetwork based on PaTch （AST） and Attention Subnetwork based on PiXel （ASX）， were proposed. These two methods were able to calculate attention within the CNN basic structure， thereby enabling bilinear sub-networks to be trained and fused in the same architecture. Experimental results on public fundus image dataset OIA-ODIR show that ABINN model has the accuracy of 85%， which is 15.8 percentage points higher than that of the common BCNN model and 0.9 percentage points higher than that of TransEye （Transformer Eye） model.

Fusing filter enhancement and reverse attention network for polyp segmentation

LIN Jianzhuang, YANG Wenzhong, TAN Sixiang, ZHOU Lexin, CHEN Danni

2023, 43(1):  265-272.  DOI: 10.11772/j.issn.1001-9081.2021111882

Asbtract ( )   HTML ( )   PDF (2283KB) ( )  

References | Related Articles | Metrics

Accurate segmentation of the polyp region in the colonoscopic images can assist doctors in diagnosing intestinal diseases. However， the structure information of polyp region is missing in the down sampling process， and the existing methods have the problems of over segmentation and under segmentation.Aiming at the problems above， a Fusing Filter enhancement and Reverse attention segmentation Network （FFRNet） was proposed. Firstly， Filter Enhancement Module （FEM） was added to the skip-connection to enhance the structure information of local lesion region in the down-sampling features. Secondly， the global features were obtained by aggregating the shallow features. Finally， Multiscale reverse Attention Fusion Mechanism （MAFM） was adopted in the up-sampling process， by combining the global features and up-sampling features to generate the reverse attention weight， the polyp region information was mined in the features layer by layer， and the relationship between the target region and the boundary was established by the guidance network to improve the integrity of the model on polyp region segmentation. On Kvasir and CVC-ClinicDB datasets， compared with Uncertainty Augmented Context Attention Network （UACANet）， FFRNet has Dice Similarity Coefficient （DSC） increased by 0.22% and 0.54% respectively. Experimental results show that FFRNet can effectively improve the accuracy of polyp image segmentation and has good generalization ability.

Improved brachial plexus nerve segmentation method based on multi-scale feature fusion

LYU Yuchao, JIANG Xi, XU Yinghao, ZHU Xijun

2023, 43(1):  273-279.  DOI: 10.11772/j.issn.1001-9081.2021111881

Asbtract ( )   HTML ( )   PDF (2862KB) ( )  

References | Related Articles | Metrics

With the features of low Signal-to-Noise Ratio （SNR） and blurred edges， ultrasound images of the brachial plexus nerve are hard to be segmented manually. Although some results have been gained by existing segmentation models， the segmentation effect is not satisfied due to the small target area and irregular shape of the brachial plexus nerve structure. Aiming at the above problems， a multi-scale feature fusion-based brachial plexus nerve segmentation model was proposed， namely Nerve-segmentation Feature Pyramid Network （Ner-FPN）. In the feature extraction stage， an Xception-like structure was designed for multi-scale feature extraction. In the prediction segmentation stage， a bidirectional FPN structure was used for feature fusion prediction. The BP （Brachial Plexus） dataset from the Kaggle brachial plexus nerve ultrasound image segmentation competition was used as the experimental data. The experimental results show that compared with the mainstream deep learning segmentation models U-Net and SegNet （Segmentation Network），the Dice Similar Coefficient （DSC） of Ner-FPN model for brachial plexus nerve segmentation can reach 0.703， which is 10.7 percentage points and 14.5 percentage points higher than those of U-Net and SegNet， and 5.5 percentage points and 3.4 percentage points higher than those of improved models QU-Net and Efficient+U-Net in the same dataset， verifying that the proposed model can be an aid for diagnosis.

Low dose CT image enhancement based on generative adversarial network

HU Ziqi, XIE Kai, WEN Chang, LI Meiran, HE Jianbiao

2023, 43(1):  280-288.  DOI: 10.11772/j.issn.1001-9081.2021101710

Asbtract ( )   HTML ( )   PDF (7479KB) ( )  

References | Related Articles | Metrics

In order to remove the noise in Low Dose Computed Tomography （LDCT） images and enhance the display effect of the denoised images， an LDCT image enhancement algorithm based on Generative Adversarial Network （GAN） was proposed. Firstly， GAN was combined with perceptual loss and structure loss to denoise the LDCT image. Then， dynamic gray?scale enhancement and edge contour enhancement were performed to the denoised image respectively. Finally， Non?Subsampled Contourlet Transform （NSCT） was used to decompose the enhanced image into multi?directional coefficient sub?images in the frequency domain， and the paired high? and low?frequency sub?images were adaptively fused with Convolutional Neural Network （CNN） to reconstruct the enhanced Computed Tomography （CT） image. Using the real clinical data of the AAPM competition as the experimental dataset， the image denoising， enhancement， and fusion experiments were carried out. The results of the proposed method are 33.015 5 dB， 0.918 5， and 5.99 on Peak Signal?to?Noise Ratio （PSNR）， Structural Similarity Index Measure （SSIM）， and Root Mean Square Error （RMSE） respectively. Experimental results show that the proposed algorithm retains the detailed information of the CT image while removing noise， and improves the brightness and contrast of the image， which helps doctors analyze the patient’s condition more accurately.

Frontier and comprehensive applications

Two-echelon location-routing model and algorithm for waste recycling considering obnoxious effect

MA Yanfang, ZHANG Wen, LI Zongmin, YAN Fang, GUO Lingyun

2023, 43(1):  289-298.  DOI: 10.11772/j.issn.1001-9081.2021111969

Asbtract ( )   HTML ( )   PDF (3080KB) ( )  

References | Related Articles | Metrics

With regard to the Location-Routing Problem （LRP） of domestic waste transfer stations and incineration stations， by considering the economic objective and the obnoxious effect of waste facilities， a piecewise function of obnoxious effect related to wind direction and distance was designed， a Two-Echelon Multi-Objective LRP （2E-MOLRP） model was formulated， and a non-dominated algorithm combining Whale Optimization Algorithm （WOA） and Simulated Annealing （SA） algorithm was proposed， namely WOA-SA. Firstly， the random method and Clarke and Wright （CW） saving algorithm were used to optimize the initial population. Secondly， a nonlinear dynamic inertia weight coefficient was adopted to adjust the convergence speed of the WOA-SA. Thirdly， the global optimization ability was enhanced by designing the parallel structure of WOA-SA. Finally， the Pareto solution set was obtained by using the non-dominated sorting method. The analysis was carried out on 35 benchmark cases such as Prins and Barreto as well as a simulated case of Tianjin. The results show that the WOA-SA can find the Best Known Solution （BKS） of 20 benchmark cases， and has the mean values of the difference between the solution results and the BKSs of 0.37% and 0.08% on Prins and Barreto cases， which proves the good convergence and stability of the WOA-SA. The proposed model and algorithm were applied to the instance， and provided three schemes with different obnoxious effect values and economic costs for decision makers with different decision preferences. Therefore， the cost of waste recycling and the obnoxious effect of facilities on environment were reduced.

Dynamic network public opinion early warning model based on improved GM（1，n）

XIE Kang, JIANG Guoqing, GUO Hangxin, LIU Zheng

2023, 43(1):  299-305.  DOI: 10.11772/j.issn.1001-9081.2021101842

Asbtract ( )   HTML ( )   PDF (1406KB) ( )  

References | Related Articles | Metrics

The free spread of public opinions may lead to the occurrence of cyber collective behaviors， which are easy to cause negative social impacts and threaten public security. Therefore， the establishment of network public opinion monitoring and early warning mechanism is necessary to prevent and control the spread of public opinions and maintain social stability. Firstly， by analyzing the formation mechanism of rumors， a prediction index system of public opinion development was constructed. Secondly， the multifactor GM（1，n） model was established to predict the development trend of the public opinion. Then， the prediction model was improved by combining with metabolism theory and Markov theory. Finally， using the “Xinjiang cotton” event and “Chengdu No.49 middle school” event in Weibo as examples， the abilities of the GM（1，n） model， the Markov GM（1，n） model and the metabolic Markov GM（1， n） model to predict the development of public opinions were compared，and the metabolic Markov GM（1， n） model was also compared with the random forest model.Experimental results show that the average prediction accuracy of the metabolic Markov GM（1， n） model is increased by 10.6% and 5.8% compared with those of the original GM（1， n） model and random forest model respectively. It can be seen that the metabolic Markov GM（1， n） model has good performance in predicting the development trend of network public opinions.

Simulation of information sharing strategy based on emergency rescue

ZHENG Wanbo, CHEN Huimin, WU Yanqing, XIA Yunni

2023, 43(1):  306-311.  DOI: 10.11772/j.issn.1001-9081.2021111988

Asbtract ( )   HTML ( )   PDF (1217KB) ( )  

References | Related Articles | Metrics

Aiming at the problem of huge losses caused by untimely and inactive emergency rescue information sharing in emergencies， a three-party game model of emergency rescue information sharing involving high-risk enterprises， rescue teams and government regulatory departments was established. Firstly， the payoff matrix and replicated dynamic equations were constructed based on the revenue. Then， stability analysis was performed for four different scenarios respectively. Finally， the evolution processes and results of the system under different scenarios were simulated through computer to obtain the optimal strategies for information sharing. Experimental results show that under the low benefit scenario， if the extra rewards and punishments are high， the willingness of emergency rescue teams to actively share rises to 0.2 and then gradually decreases until they reject information sharing completely； if the extra cost is high， the willingness of high-risk enterprises to actively share rises to about 0.2 and then rapidly decreases to 0. Meanwhile， the behavioral strategies of participants are most sensitive to the changes of positive benefit， then to the changes in extra rewards and punishments and extra costs. The above results can provide guidance for the selection of information sharing strategies in emergency response.

Walking control and stability analysis of flexible biped robot with variable length legs

LIAO Fakang, ZHOU Yali, ZHANG Qizhi

2023, 43(1):  312-320.  DOI: 10.11772/j.issn.1001-9081.2021111953

Asbtract ( )   HTML ( )   PDF (4678KB) ( )  

References | Related Articles | Metrics

Aiming at the problem that the traditional biped robot model lacks the feet mass and the torso， a flexible biped robot model considering the influence of swing leg dynamics and torso was proposed， and its walking control and stability were studied. Firstly， the dynamics model of the system was established and the dynamics equation was deduced by the Euler-Lagrange method. At the same time， based on the Spring-Loaded Inverted Pendulum （SLIP） model， by adding rigid torso， foot mass， and adopting telescopic legs of variable length， the influence of the torso and the dynamics of swing legs on the gait of the robot was fully considered. Then， the feedback linearization controller based on variable length legs was designed to track the target trajectory and regulate the attitudes of the swing legs and the torso. Finally， the Newton-Raphson iteration method and Poincaré map were adopted to analyze the fixed point and orbital stability conditions of the robot. Simulation analysis was carried out based on theoretical analysis. Simulation results show that the proposed controller can realize the robot’s periodic walking and has good robustness to the external interference. And the moduli of all eigenvalues of the Jacobian matrix are less than 1， forming a stable limit cycle， which proves that the system has orbital stability.

Deep learning model for multi-station temperature prediction combined with MOD11A1 and surface meteorological station data

ZHANG Jun, WU Pengli, SHI Lukui, SHI Jin, PAN Bin

2023, 43(1):  321-328.  DOI: 10.11772/j.issn.1001-9081.2021111888

Asbtract ( )   HTML ( )   PDF (3429KB) ( )  

References | Related Articles | Metrics

Focusing on the issues that the relationships between the stations are affected by the sparse distribution of surface meteorological stations and it is difficult to infer the strengths of relationships between the stations， a Deep learning Model for multi-station temperature prediction combined with MOD11A1 and surface meteorological station data was proposed， namely GDM， which included Spatio-Temporal Attention （TSA） ， Double Graph neural Long Short-Term Memory （DG-LSTM） network encoding and Edge-Node transform Gated Recurrent Unit （EN-GRU） decoding modules. Firstly， TSA module was utilized to extract MOD11A1 image features and form the temperature time series of multiple virtual meteorological stations， so as to alleviate the impact of sparse distribution of surface meteorological stations on the relationships between the stations. Secondly， DG-LSTM encoder was used to calculate the strengths of the relationships among surface meteorological stations and virtual meteorological stations via fusing two sets of temperature time series. Finally， EN-GRU decoder was adopted to model the temperature time series relationships between surface meteorological stations through combining the inter-station relationship strengths. Experimental results show that compared with 2-Dimensional Convolutional Neural Network （2D-CNN）， Long Short-Term Memory-Fully Connected network （LSTM-FC）， Long Short-Term Memory neural network Extended （LSTME） and Long Short-Term Memory and AdaBoost network （LSTM-AdaBoost）， GDM has the Average Absolute Error （MAE） of temperature prediction in 24 hours at 10 surface meteorological stations reduced by 0.383 ℃， 0.184 ℃， 0.178 ℃ and 0.164 ℃ respectively. It can be seen that GDM can improve the prediction accuracy of the temperature for meteorological stations in the next 24 hours.