In-Vehicle Network Injection Attacks Detection Based on Feature Selection and Classification

作     者：Haojie Ji Liyong Wang Hongmao Qin Yinghui Wang Junjie Zhang Biao Chen 

作者机构：Key Laboratory of Modern Measurement and Control TechnologyMinistry of EducationBeijing Information Science and Technology UniversityBeijing 100192China State Key Laboratory of Advanced Design and Manufacturing for Vehicle BodyCollege of Mechanical and Vehicle EngineeringHunan UniversityChangsha 410082China School of Transportation Science and EngineeringBeihang UniversityBeijing 100191China Hefei Innovation Research InstituteBeihang UniversityHefei 230012China 

出 版 物：《Automotive Innovation》 (汽车创新工程（英文）)

年 卷 期：2024年第7卷第1期

页      面：138-149页

核心收录：

学科分类：0808[工学-电气工程] 08[工学] 080204[工学-车辆工程] 0802[工学-机械工程] 0823[工学-交通运输工程] 

基　　金：supported by the the Young Scientists Fund of the National Natural Science Foundation of China under Grant 52102447 by the Research Fund Project of Beijing Information Science&Technology University under Grant 2023XJJ33. 

主　　题：Classification algorithm Anomaly detection In-vehicle network Feature extraction Injecting attack 

摘      要：Detecting abnormal data generated from cyberattacks has emerged as a crucial approach for identifying security threats within in-vehicle networks.The transmission of information through in-vehicle networks needs to follow specific data for-mats and communication protocols regulations.Typically,statistical algorithms are employed to learn these variation rules and facilitate the identification of abnormal data.However,the effectiveness of anomaly detection outcomes often falls short when confronted with highly deceptive in-vehicle network attacks.In this study,seven representative classification algorithms are selected to detect common in-vehicle network attacks,and a comparative analysis is employed to identify the most suitable and favorable detection method.In consideration of the communication protocol characteristics of in-vehicle networks,an optimal convolutional neural network(CNN)detection algorithm is proposed that uses data field characteristics and classifier selection,and its comprehensive performance is tested.In addition,the concept of Hamming distance between two adjacent packets within the in-vehicle network is introduced,enabling the proposal of an enhanced CNN algorithm that achieves robust detection of challenging-to-identify abnormal data.This paper also presents the proposed CNN classifica-tion algorithm that effectively addresses the issue of high false negative rate(FNR)in abnormal data detection based on the timestamp feature of data packets.The experimental results validate the efficacy of the proposed abnormal data detection algorithm,highlighting its strong detection performance and its potential to provide an effective solution for safeguarding the security of in-vehicle network information.