Intrusion Detection System for Vehicular Networks Based on MobileNetV3

With the advancement and refinement of intelligence and connectivity, intelligent connected vehicles have emerged as a prominent trend in contemporary development. Consequently, invasion attacks targeting these intelligent connected vehicles have also arisen. Mainstream intrusion detection systems (IDS) based on deep learning technologies can address malicious traffic infiltrations; however, they often fail to meet the real-time and lightweight requirements of vehicles. This paper introduces a lightweight vehicular intrusion detection method leveraging the MobileNetV3 architecture. By utilizing MobileNetV3 as the core framework, this method incorporates advanced techniques and design principles such as Depthwise Separable Convolution, Bottleneck structures, and Squeeze-and-Excitation (SE) modules. These innovations significantly reduce computational and parameter overhead while maintaining high model accuracy. Furthermore, MobileNetV3 is specifically designed for deployment on mobile devices, ensuring efficient operation even in resource-constrained environments. The proposed intrusion detection model achieved an accuracy, recall, precision, and F1 score of 100% on the Car-Hacking dataset, and an accuracy, recall, precision, and F1 score of 99.98% on the CICIDS-2017 dataset. The model size is 16MB. Experimental results demonstrate that this intrusion detection scheme not only accurately detects malicious attacks on vehicles but also meets the lightweight requirements of vehicular applications.