Improving Real-Time Detection of Abnormal Traffic Using MobileNetV3 in a Cloud Environment
Yihuan Mao, Wei Fu, Yue Zhao, Jinhong Chen · Electronics · 2025
To address the dual challenges of achieving high classification accuracy and real-time performance when detecting abnormal traffic in cloud environments, this study introduces an improved lightweight real-time detection model, IM-MobileNetV3. By standardizing features, dividing time windows, mapping spatially, and reshaping images, the model converts traffic features into 224 × 224 × 3RGB images, enabling multi-angle representation of traffic features and adapting to the model’s input requirements. Additionally, the MobileNet-Small model is improved by introducing the ECA module to replace the original SE module, avoiding the information loss that is often caused by channel dimension reduction, and the initial convolutional layer, bottleneck layer, and tail structure are optimized to enhance the feature extraction capability. The experimental results show that, on the CIC-IDS2018 dataset, IM-MobileNetV3 achieves an accuracy rate of 96.9%, reduces the number of parameters by 33% compared to the original model, and has a single-sample inference time of only 2.8 ms, significantly outperforming mainstream models.