融合改进TCN与DRSN的IoT入侵检测模型  

IoT Intrusion Detection Model Integrating Improved TCN and DRSN

作　　者：赵建 姜伟[1] ZHAO Jian;JIANG Wei(School of Computer Science and Information Engineering,Harbin Normal University,Harbin 150025,China)

机构地区：[1]哈尔滨师范大学计算机科学与信息工程学院,哈尔滨150025

出　　处：《小型微型计算机系统》2025年第2期474-481,共8页Journal of Chinese Computer Systems

基　　金：国家自然科学基金面上项目(61872104)资助;黑龙江省教育厅科研项目(11551124)资助;哈尔滨师范大学研究生创新工程项目(HSDSSCX2023-10)资助。

摘　　要：入侵检测系统已逐步成为物联网安全的重要防护手段.然而,现有物联网入侵检测模型的样本数据存在类别不平衡、特征提取不足等问题,这导致了对于小类别攻击的低识别率与较低的精确率.因此,本文提出了一种融合改进时域卷积网络与深度残差收缩网络的物联网入侵检测模型.首先,利用扩张因果卷积与一维卷积充分提取数据的时空特征,形成深层层次的网络结构;然后引入自我注意的软门槛,能够无需专家经验自动地设置门槛,消除冗余特征;最后,使用焦点损失函数来增强对少数类的识别率.实验在TON-IoT数据集上的总体准确率和F1值分别高达99.88%和99.64%,其中小样本类的F1值为100%.实验结果表明,与其他模型相比,所提模型显著提高了对于不平衡入侵数据的检测能力.Intrusion detection systems have gradually become an important means of protection for IoT security.However,the sample data of existing IoT intrusion detection models have problems such as category imbalance and insufficient feature extraction,which results in low recognition rates and relatively low recognition rates for small category attacks.Therefore,this paper proposes an IoT intrusion detection model that integrates an improved temporal convolutional network and a deep residual shrinkage network.First,dilated causal convolution and one-dimensional convolution are used to fully extract the spatiotemporal features of the data,forming a deep-level network structure;then introduce a soft threshold of self-attention,which can automatically set the threshold without expert experience and eliminate redundant features;finally,use the focus loss function to enhance the recognition rate of minority classes.The experiment is in TON-IoT The overall accuracy and F1 value on the data set are as high as 99.88%and 99.64%respectively,among which the F1 value of the small sample class is 100%.Experimental results show that compared with other models,the proposed model significantly improves the detection ability of imbalanced intrusion data.

关 键 词：物联网 入侵检测 时域卷积网络 深度残差收缩网络 样本不平衡 焦点损失函数 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]