Table of Content

30 April 2024, Volume 0 Issue 04

Previous Issue   

A mmWave Massive MIMO Channel Estimation Based on Joint Weighted#br# and Truncated Nuclear Norm

ZHANG Zhineng, HUANG Xuejun

2024, 0(04):  1-4.  doi:10.3969/j.issn.1006-2475.2024.04.001

Asbtract ( 240 )   PDF (1413KB) ( 206 )  

References | Related Articles | Metrics

Abstract： In this paper， a millimeter-wave massive multiple input multiple output （MIMO） channel estimation algorithm based on joint weighted and truncated nuclear norm is proposed. Aiming at the problem of high training and feedback overhead in millimeter-wave massive MIMO channel estimation， firstly， the channel estimation problem is transformed into a low-rank matrix recovery problem by using the sparse antenna angle domain of millimeter-wave channel. An effective and flexible rank function， the joint weighted and truncated kernel norm， is adopted as the relaxation of the nuclear norm， and a new matrix recovery model is constructed for channel estimation. The optimization objective is to minimize the weighted and truncated nuclear norm， and it is solved by an alternating optimization framework. The simulation results show that this method can effectively improve the accuracy of channel estimation and has reliable convergence.

Federated Learning Aggregation Algorithm Based on AP Clustering Algorithm

AO Bochao, FAN Bingbing

2024, 0(04):  5-11.  doi:10.3969/j.issn.1006-2475.2024.04.002

Asbtract ( 224 )   PDF (1628KB) ( 167 )  

References | Related Articles | Metrics


Abstract: In traditional federation learning， multiple clients’ local models are trained independently from their private data， and the central server generates a shared global model by aggregating the local models. However， due to statistical heterogeneity such as non-independent identically distributed （Non-IID） data， a global model often cannot be adapted to each client. To address this problem， this paper proposes an AP clustering algorithm-based federation learning aggregation algorithm （APFL） for Non-IID data. In APFL， the server calculates the similarity matrix between each client based on the data characteristics of the clients， and then uses the AP clustering algorithm to divide the clients into different clusters and construct a polycentric framework to calculate the suitable personalized model weights for each client. This algorithm is experimented on FMINST dataset and CIFAR10 dataset， and APFL improves 1.88 percentage points on FMNIST dataset and 6.08 percentage points on CIFAR10 dataset compared with traditional Federated Learning FedAvg. The results show that the proposed APFL improves the accuracy performance of Federated Learning on Non-IID data in this paper.


Review and Discussion of Personalised News Recommendation Systems

ZHAI Mei

2024, 0(04):  12-20.  doi:10.3969/j.issn.1006-2475.2024.04.003

Asbtract ( 610 )   PDF (1534KB) ( 352 )  

References | Related Articles | Metrics


Abstract： With the rapid development of news media technology and the exponential growth of the number of online news， personalised news recommendation plays an extremely crucial role in order to solve the problem of online information overload. It learns users' browsing behaviour， interests and other information， and actively provides user with news of interest， thus improving user's reading experience. Personalised news recommendation has become a hot research and practical problem in the field of journalism and computer science， and experts in the industry have proposed various recommendation algorithms to improve the performance of recommendation systems. In this paper， we systematically describe the latest research status and progress of personalised news recommendation. firstly， we briefly introduce the architecture of news recommendation systems， and then we study the key recommendation algorithms and common evaluation metrics in news recommendation systems. Although personalised news recommendation brings a good experience to users， it also brings a lot of unknown effects to users. Unlike other news recommendation reviews， this paper also examines the impact of current news recommendation systems on user behaviour and the problems they face. Finally， the paper proposes research directions and future work on personalised news recommendation based on the current problems encountered.



Multi-agent Genetic Algorithm Based Cloud Platform Anti-fake Data Injection Attack Method

WANG Dongyue, LIU Hao

2024, 0(04):  21-26.  doi:10.3969/j.issn.1006-2475.2024.04.004

Asbtract ( 234 )   PDF (844KB) ( 120 )  

References | Related Articles | Metrics

Abstract： In order to ensure the security of data transmission in cloud platforms， a multi-agent genetic algorithm based anti false data injection attack method for cloud platforms is proposed. We build a cloud platform using the open source platform OpenStack， and analyze the process of false data injection attacks on the cloud platform. Based on this attack process， a false data injection attack detection framework is constructed by combining Copula function and GAN generation countermeasures network. The discriminator and generator in the Copula GAN function model are used to conduct countermeasures training on the original measured data of the cloud platform， and then an extreme random tree classifier is used to detect false data to determine whether there is a false data injection attack in the cloud platform. Using a three-layer attack and defense game model to defend against false data injection attacks in the cloud platform， the model allocates defense resources for each data transmission line， and sets corresponding constraints. The model is optimized using a multi-agent genetic algorithm to complete the defense against false data injection attacks on cloud platforms. The experimental results show that this method can accurately detect false data on cloud platforms and take timely defensive measures， and has a strong ability to resist false data injection attacks.



Method of Location and Capacity Determination for Distributed Generator Based on#br# Enhanced Capuchin Search Algorithm

LI Jiaduo, YAN Xiuying

2024, 0(04):  27-32.  doi:10.3969/j.issn.1006-2475.2024.04.005

Asbtract ( 199 )   PDF (1710KB) ( 77 )  

References | Related Articles | Metrics


Abstract： Under the influence of the characteristics and distribution of new energy， large-scale distributed generator （DG） and electric vehicles （EV） are connected to the distributed network， which makes a fundamental change in the structure and operation mode of the power grid. In order to improve the utilization rate of DG and reduce the fluctuation of the distributed network， the minimum annual comprehensive cost of the distribution generation is taken as the objective function. Under the constraint conditions of node voltage and branch power， a planning model of constant volume location of distributed network DG is established including EV charging， and an enhanced capuchin search algorithm is proposed to solve the model. This algorithm improves the social behaviors selected by the leader in the wild horse optimizer on the traditional capuchin search algorithm to avoid falling into the local optimal situation. Finally， the typical distribution system of IEEE-33 bus is simulated and compared with other algorithms to verify the superiority of the proposed algorithm.



A Method of Predicting Open Flow Potential

MENG Yalei1, SHI Hongyu1, WANG Yu2

2024, 0(04):  33-37.  doi:10.3969/j.issn.1006-2475.2024.04.006

Asbtract ( 235 )   PDF (810KB) ( 105 )  

References | Related Articles | Metrics


Abstract： The traditional method of predicting open flow potential is calculated through gas test operation， which has long operation cycle and large capital investment. However， in the early stage of gas exploration and development， a large number of parameters obtained from logging can effectively reflect the advantages and disadvantages of reservoir properties. In this paper， a relationship model between logging parameters and open flow potential is established. First， K-means clustering analysis method is used to analyze and select， preprocess and transform logging parameters. Then， the improved ID3 algorithm is used to design the open flow potential prediction method， to construct the open flow potential decision tree and form the open flow potential decision scheme for the target interval of a single well. The experiment results show that this method can effectively predict open flow potential and single well productivity by using logging parameters. Through omitting the gas test operation， we can speed up the gas field production schedule， reduce the development cost， and improve the economic benefits of gas field development.


GAN-generated Fake Images Recognition Based on Improved ConvNeXt

XIAO Mengsi1, WU Jianbin1, TU Yameng1, YUAN Linfeng2

2024, 0(04):  38-42.  doi:10.3969/j.issn.1006-2475.2024.04.007

Asbtract ( 345 )   PDF (3317KB) ( 166 )  

References | Related Articles | Metrics


Abstract: In order to distinguish the authenticity of face images in social networks， a recognition method based on ConvNeXt for face image generated by Generative adversarial networks （GAN） is proposed. The ConvNeXt network structure is used as the main body， using the color features and spatial texture features of the face image， and multi-channel combination input （Multichannel Input， MCI） with multi-color space is used to expand the learning range of the network， while channel attention mechanism and spatial attention mechanism are introduced to highlight the differences between real and fake face images in color components and spatial features， and then the detection and recognition of fake face images are achieved. The experimental results show that the recognition accuracy of face images generated by GAN with improved ConvNeXt （I-ConvNeXt） network structure reaches 99.405%， with an average accuracy improvement of 1.455 percentage points compared with the original ConvNeXt algorithm. The results validate the feasibility and reasonableness of the proposed scheme.



Ghost Convolution Based Prediction Method for Recurrence of High Grade#br# Serous Ovarian Cancer

TANG Yibo1, CUI Shaoguo1, WAN Haoming1, WANG Rui1, LIU Lili2

2024, 0(04):  43-47.  doi:10.3969/j.issn.1006-2475.2024.04.008

Asbtract ( 199 )   PDF (1343KB) ( 120 )  

References | Related Articles | Metrics


Abstract： High grade serous ovarian cancer is a malignant tumor disease， and preoperative recurrence prediction can help clinical doctors provide personalized treatment plans for patients and reduce the mortality rate. Due to the less and difficult-to-obtain medical data of this disease， its deep learning model is difficult to obtain sufficient training， and the accuracy of recurrence prediction needs to be improved. To address this issue， this article proposes an improved low-parameter residual network TGE-ResNet34， which uses ResNet34 as the backbone network and replaces traditional convolution modules with Ghost convolutions to extract lesion area features and reduce the model’s parameter volume. The ECA （Efficient Channel Attention） attention mechanism is incorporated between two Ghost convolutions to suppress interference from useless feature extraction. Finally， the model is evaluated through a five-fold cross-validation to avoid the randomness of data partitioning. The experimental results show that the accuracy of the improved TGE-ResNet34 network is 96.01%， which is 4.52 percentage points higher than the original baseline network’s accuracy and reduces the parameter volume by 15.98 M.



Tongue Image Segmentation Algorithm Based on Dilated ADU-Net in Open Environment#br# #br#

WANG Xin, XIN Guojiang, ZHANG Yang, ZHU Lei ​

2024, 0(04):  48-54.  doi:10.3969/j.issn.1006-2475.2024.04.009

Asbtract ( 243 )   PDF (2095KB) ( 190 )  

References | Related Articles | Metrics


Abstract: Accurate tongue image segmentation is an important prerequisite for obtaining correct tongue diagnosis results. Aiming at the problem that traditional segmentation algorithms are difficult to accurately and stably segment tongue images under complex lighting conditions， an improved U-Net tongue image segmentation model （Dilated Attention & Dense U-Net， Dilated ADU-Net） combining dilated convolution dual attention mechanism and dense connection mechanism is constructed. Firstly， the backbone network is built based on the symmetric structure of U-Net network. Then， the downsampling module uses a cavity mixed attention module to make the network focus on tongue features， and the upsampling module uses a dense connection mechanism to fuse multi-layer feature information. Finally， the tongue image dataset in open environment is used to train the network to obtain the tongue image segmentation model. Experimental verification shows that compared with other advanced segmentation methods， the mean Intersection over Union （mIoU） of tongue image segmentation model constructed in this paper reaches 96.73% and the similarity coefficient Dice （DSC） reaches 98.08%， which has better segmentation performance and can realize accurate segmentation of tongue image in complex environments.

Helmet Detection Algorithm Based on CE-YOLOv5s

WANG Zhibo, MA Han, FENG Jinliang, LIU Guoming

2024, 0(04):  55-59.  doi:10.3969/j.issn.1006-2475.2024.04.010

Asbtract ( 233 )   PDF (2091KB) ( 178 )  

References | Related Articles | Metrics


Abstract： In the complex environment of construction sites， there are many dangerous factors， so the protection of the safety of workers has become a focus. Due to the chaotic environment and fixed information collection points at construction sites， there are problems of missed and false detection in safety helmet-wearing detection. Therefore， this paper proposes a safety helmet detection algorithm based on CE-YOLOv5s. The algorithm combines the SE attention mechanism with the C3 module， replaces the C3 module in the original network， assigns a higher weight to key features， and suppresses general features. Meanwhile， an object detection neural network based on Bi-directional Feature Pyramid Network （BiFPN） is introduced， which performs both upward and downward feature fusion， adds additional weights to each channel， and better preserves detailed information under low-resolution images. The SIoU loss function is introduced to improve the accuracy of boundary box positioning and accelerate convergence speed. Experimental results show that the improved network model has significantly improved in precision， recall， mAP@0.5， and mAP@0.5：0.95， effectively improving the detection accuracy of safety helmets and improving the detection accuracy of small targets and obscured targets in cluttered backgrounds. When applied to construction sites， it can timely detect whether workers have taken protective measures， and better protect their safety.



Supermarket Fruit and Vegetable Retrieval Method Based on Deep Learning

GUO Zexin, ZHONG Guoyun, HE Jianfeng, ZHANG Jun

2024, 0(04):  60-65.  doi:10.3969/j.issn.1006-2475.2024.04.011

Asbtract ( 203 )   PDF (1335KB) ( 116 )  

References | Related Articles | Metrics


Abstract: In view of the problems that the current settlement method of supermarket fruits and vegetables cannot add new categories and low accuracy of small sample recognition， this paper proposes a supermarket fruits and vegetables retrieval method based on deep learning. The method obtains fruit and vegetable subjects through YOLOv4 to remove redundant background information， and extracts corresponding deep semantic features of fruit and vegetable subjects through MobileNetV3. Finally， category judgment is completed according to metric learning technology. This paper conducts experiments in accordance with the actual operation conditions of supermarkets and concludes that the method could accurately identify different fruit and vegetable categories under the condition of small samples. When the number of samples for each category is 15， the average recognition rate is about 94%， the time cost is 0.93s， and the new categories could be updated in real time. This method greatly reduces the huge labor and time cost in the actual operation of traditional supermarkets， and provides a solution for the realization of intelligence and automation in the fruit and vegetable retail industry.



Short Text Classification Method Based on Improved Adversarial Learning and Fusion Features

NING Zhaoyang1, 2, SHEN Qing2, 3, HAO Xiulan1, 2, ZHAO Kang1, 2

2024, 0(04):  66-76.  doi:10.3969/j.issn.1006-2475.2024.04.012

Asbtract ( 185 )   PDF (1883KB) ( 221 )  

References | Related Articles | Metrics

Abstract: Text classification is one of the most important directions in natural language processing research. Short text has the characteristics of less word count， ambiguity， less key information and not easy to capture， but the algorithm models that classify them are often different in training and reasoning， and mainstream classification models basically model key features and ignore non-key feature information， which increases the challenges in accurate classification. In order to solve the above problems， this paper proposes a short text classification framework combining the fusion of multiple adversarial training strategies and improving the self-attention mechanism. At the beginning， the model adds adversarial perturbation to the text vector representation level to strengthen the text representation ability， and adds an adversarial perturbation to improve the model weights after the F1 score reaches a certain threshold to strengthen the generalization ability of the model during training and inference， thereby assisting in improving the feature learning ability of each classifier of the framework. In terms of feature learning network module， this paper uses the combination of multi-scale convolutional module and bidirectional long short-term memory neural network to learn different granular features， in order to learn nonadjacent feature information， introduces hole convolution， increases the convolution receptive field， and designs a gating mechanism to control the learning speed of this layer information. Finally， by adding a new attention mechanism， the key information is modeled and the non-critical information is modeled， and the loss is added for calculation， which enhances the model’s ability to learn feature information and reduces the risk of overfitting. The tests of THUCNews news headline dataset and Toutiao headline dataset of two large-scale public datasets show that the F1 score of this method is increased by up to 4.93 percentage points and 6.14 percentage points compared with the current mainstream model and classical model， and the effectiveness of adding weight disturbance threshold and different modules is also compared and ablation experiments are explored.


Features Analysis of Suicide Ideation Causes Based on Machine Learning

FU Qi1, ZHANG Liyuan2, DAI Huan3

2024, 0(04):  77-82.  doi:10.3969/j.issn.1006-2475.2024.04.013

Asbtract ( 166 )   PDF (742KB) ( 328 )  

References | Related Articles | Metrics

Abstract： Suicide is one of the most significant public health crises globally， surpassing the combined mortality rate of wars， homicides， and natural disasters. This study employs computer technology， machine learning， and deep learning methods to analyze social media texts that contain suicidal ideation， aiming to automatically extract the underlying causes of suicidal thoughts. The study investigates the impact of content features (such as words， parts of speech， dependency syntactic parsing) and emotional-psychological features (including linguistics， emotions， suicidal psychology) on the task of automatically extracting causes of suicidal ideation. Experimental results indicate that content features perform notably well and are the most significant and crucial factors among the features. Specifically， word features exhibit the best performance， while parts of speech and dependency syntactic parsing features are overshadowed by the inclusion of word features to some extent. In contrast， emotional-psychological features effectively complement and enhance content features. The expression of emotions， sentiments， or psychological aspects shows a positive correlation with the underlying causes of suicidal ideation.



Encryption Traffic Classification Method Based on AHP-CNN

YOU Jiajing1, 2, HE Yueshun1, HE Linlin1, ZHONG Hailong1, 2

2024, 0(04):  83-87.  doi:10.3969/j.issn.1006-2475.2024.04.014

Asbtract ( 285 )   PDF (895KB) ( 236 )  

References | Related Articles | Metrics

Abstract: To address the insufficient feature extraction of existing methods for encrypted traffic， this study proposes an encrypted traffic classification method based on an Attention-based Hybrid Pooling Convolutional Neural Network （AHP-CNN）. This method improves the pooling layers of Convolutional Neural Networks （CNNs） by combining average pooling and max pooling in a parallel manner， forming a dual-layer synchronized pooling pattern. This enables the capturing of both global and local features of network encrypted traffic. Furthermore， a self-attention module is incorporated into the model to enhance the extraction of dependency relationships among encrypted traffic features， leading to more accurate classification. Experimental results demonstrate a significant improvement in the accuracy of encrypted traffic identification using the proposed model， with an F1 score exceeding 0.94. This research provides a more effective and precise approach for the classification of network encrypted traffic， contributing to advancements in research and applications in the field of network security.

A Source Code Security Vulnerability Detection Method Using ChatGPT

YU Lihui, HU Shaowen, HUANG Langxin, LUO Shuhuan

2024, 0(04):  88-91.  doi:10.3969/j.issn.1006-2475.2024.04.015

Asbtract ( 258 )   PDF (3675KB) ( 206 )  

References | Related Articles | Metrics

Abstract: As the security issues of software and information systems become more and more prominent, as an important part, the security of source code is the bottom key point. How to quickly and accurately detect security vulnerabilities of source code is particularly important. This paper proposes a source code security vulnerability detection method based on ChatGPT, which takes advantage of ChatGPT in the field of natural language processing, converts source code into natural language form, and then uses ChatGPT to process it to identify potential security vulnerabilities. This method can detect various types of security vulnerabilities, such as insecure design, SQL injection and so on. We demonstrate the superiority and accuracy of our approach through experimental analysis of security vulnerability detection on source codes of publicly available datasets.



Improved Mayfly Algorithm for Integrated of Process Planning and Scheduling

YANG Ke1, PAN Dazhi1, 2, CHI Ying1

2024, 0(04):  92-98.  doi:10.3969/j.issn.1006-2475.2024.04.016

Asbtract ( 171 )   PDF (2037KB) ( 107 )  

References | Related Articles | Metrics

Abstract: Aiming at the integrated of process planning and scheduling （IPPS）， a mathematical model is established based on AND/OR disjunction graph and the concept of “combination” to minimize the makespan optimization target. An integer coding scheme is designed， which does not need to generate a processing path for the workpiece in advance， can deal with process planning and scheduling problems at the same time. Due to the effect of the initial solution on the algorithm’s ability to find the optimal， the load of the machine is considered to improve the quality of the initial population when the population is initialized. The plug-in method is used to generate active scheduling during decoding to shorten the overall processing time. The discrete mayfly algorithm is used to solve IPPS， and the Metropolis criterion is used to receive poor solutions， and the adaptive factors are introduced to improve the convergence speed of the algorithm， and a local search algorithm is designed to improve the accuracy of the algorithm. Finally， large-scale benchmark studies are carried out and compared with other algorithms to verify that the algorithm has good optimization performance.


Decentralized Federation Learning Based on Fed-DPDOBO

YANG Ju, DENG Zhiliang, YANG Zhiqiang, WANG Yan, ZHAO Zhongyuan

2024, 0(04):  99-106.  doi:10.3969/j.issn.1006-2475.2024.04.017

Asbtract ( 185 )   PDF (1927KB) ( 137 )  

References | Related Articles | Metrics


Abstract： The traditional client-server architecture federation learning is an effective means of solving the problem of data silos， where the central server is under enormous bandwidth pressure and the decentralized peer-to-peer architecture federation learning improves this situation to some extent. However， clients of federal learning also suffer from the risk of data privacy breaches and the gradient information of their cost function is difficult to obtain in some cases. To address these issues， this paper designs an Federated Differential Privacy Distributed One-point Bandit Online algorithm （Fed-DPDOBO） for peer-to-peer architecture federation learning under consistency constraints， which effectively addresses the problems of bandwidth limitation of the central node and unknown gradient information of the client. In addition， data privacy for each client is well protected due to the use of differential privacy technology. Finally， the effectiveness of this paper's algorithm is verified by conducting decentralized federation learning experiments with the MINST dataset.



Structural Attention Mechanism Auto-encoder for miRNA-disease Association Prediction

XIE Guobo, LUO Canjie, LIN Zhiyi, JIANG Zelin

2024, 0(04):  107-114.  doi:10.3969/j.issn.1006-2475.2024.04.018

Asbtract ( 170 )   PDF (1211KB) ( 96 )  

References | Related Articles | Metrics


Abstract： Research on MicroRNA （miRNA） -disease association prediction is helpful for human disease prevention， diagnosis and treatment， etc. Many researchers have developed miRNA-disease association prediction methods based on graph auto-encoder. However， most of the auto-encoder methods do not consider the difference between neighbor nodes when coding the central node. Therefore， this paper proposes a new method called structural attention mechanism auto-encoder for miRNA-disease association prediction （SAAE）. The SAAE model uses an encoder based on graph neural network， which uses multiple coding layers to explore the information of multi-order neighbors. In order to fuse the feature information of the central node and the neighbor node with different weights and capture the structure information of the node in the graph， the structured attention mechanism is introduced in the coding layer to encode the original information of the graph node to generate new feature information. Subsequently， the decoding is performed by a decoder， and the decoded feature information is mined for potential associations between miRNA and disease nodes using a random forest algorithm. The experimental results show that the average area under the curve of SAAE under five-fold cross validation is 94.53%. In addition， two case studies on kidney tumors and lungtumors were conducted to verify the validity of SAAE prediction.

iOS Application Development Framework Based on YAML

FAN Liangjun1, PENG Zhenwan1, WANG Chen2, YU Hongtao2, LIANG Zhen1

2024, 0(04):  115-120.  doi:10.3969/j.issn.1006-2475.2024.04.019

Asbtract ( 161 )   PDF (1058KB) ( 114 )  

References | Related Articles | Metrics

Abstract: To solve the problems of difficulty for beginners to write iOS programs using Swift language and the high cost of developing multi platform applications for the company， this article implements an iOS framework for Apple application development based on the concept of cross platform development， which involves writing a simple and easy to learn explanatory configuration file YAML. Firstly， the feasibility of developing Apple applications using third-party languages is investigated. Secondly， UI controls， style rendering， and network requests are abstracted and structurally defined in YAML syntax. The underlying framework completes core functions such as resource loading， data processing， control creation， and user interaction in sequence based on the structural definition. And based on this framework， an online consultation application （Aimed Care） is designed and developed， and successfully launched on the Apple Store. Coupled with the Web and Android frameworks， the goal of developing a set of code for publishing on multiple platforms can be achieved. The successful release of the application indicates that the framework meets the development standards of iOS software from the perspectives of functionality and security， and can be applied to iOS development.



A Marginal End Resource Distribution Scheme Based on Meta-universe

ZHAO Chenyi1, ZHAO Xin2

2024, 0(04):  121-126.  doi:10.3969/j.issn.1006-2475.2024.04.020

Asbtract ( 165 )   PDF (1232KB) ( 83 )  

References | Related Articles | Metrics

Abstract: In order to satisfy the quality of service and minimize the cost of resource utilization in the meta-universe， this paper proposes an edge end resource distribution scheme. To be specific， this paper first introduces the architecture and application of the meta-universe briefly. Then， taking the education industry as an example， based on the two-stage stochastic integer programming method， the paper studies the optimal resource distribution scheme at the edge end to minimize the cost of virtual service providers. The performance of the proposed method is evaluated by numerical research and experimental simulation. Compared with the benchmark method， the proposed method can better adapt to the change of user demand probability and minimize the business cost.