基于CBAMTL-MobileNet V3的车载网络入侵检测

Vehicle Network Intrusion Detection Based on CBAMTL-MobileNet V3

提出了一种基于CBAMTL-MobileNet V3的车载网络入侵检测方法。该方法使用轻量级模型MobileNet V3,减少其层数加快模型的训练和检测速度;将模型中的SE模块置换为注意力模块(CBAM)使模型更聚焦于特定特征,提高特征提取能力,进而提高检测攻击的精确度;引入迁移学习对模型权重进行微调,减少参数和内存资源的消耗,缩短了训练时间,使模型表现出更快的运算速度。仿真结果表明:所提模型的各项检测指标都优于MobileNet V3模型。与其他模型相比,所提模型既具备轻量级模型的高效性,同时又高于其他复杂模型的检测精度,识别各类别攻击的性能最优。

A vehicle network intrusion detection method is proposed based on CBAMTL-MobileNet V3.The lightweight MobileNet V3 model is used,and reduced its layers to improve both training and detection speeds.The squeeze-and-excitation(SE)modules in the model are replaced with convolutional block attention module(CBAM)to focus the model more on specific features,enhancing feature extraction capabilities and consequently improving the accuracy of attack detection.Transfer learning is introduced to fine-tune the model weights,reducing parameter and memory resource consumption,thereby shortening the training time and improving the computational speed of the model.Simulation results indicate that the proposed model is better than the MobileNet V3 model in various detection indexes.Compared with other models,the proposed model exhibits both the efficiency of a lightweight model and higher detection accuracy than other complex models,making it the optimal performer in recognizing various types of attacks.