Abstract
The ever-evolving nature of cyber-attacks within the Internet of Things (IoT) highlights the urgent need for developing advanced detection methods suitable for the limited resources of edge environments. Traditional detection methods face challenges in adapting to the rapidly changing and expanding landscape of IoT applications. This situation underscores the necessity for advanced detection methods that can achieve high detection accuracy while minimizing resource consumption. To address this issue, this paper proposes a lightweight framework for intrusion detection that combines the strengths of lightweight vision transformers with transfer learning. By incorporating a pre-trained MobileViT model into the domain of IoT intrusion detection, the proposed framework utilizes transfer learning to efficiently extract features and classify data, thereby enhancing detection performance. The approach involves converting traffic data into images by segmenting the data into blocks of successive samples and converting these blocks into grayscale images. Experimental results demonstrate the proposed framework's superior performance, achieving a high accuracy of 99.97% and an F1 score of 99.92% in a multi-class classification scenario on the Edge-IIoTset dataset, outperforming existing methods.