Lightweight IoT Device Recognition Model based on Statistical and Sequential Feature Fusion

The extensive use of Internet of Things (IoT) devices, along with limited resources and significant variability in their functions, has resulted in problems like identity theft and unauthorized access in IoT security management. Device fingerprints construction for the effective identification of device IDs has emerged as a prominent and challenging research subject. This study aims to tackle the disparity between classification efficiency and accuracy, which arise from inadequate representation of individual features and the difficulties in integrating mixed features during the recognition of IoT devices. An authentication model concerning traffic statistical features is proposed. It establishes a dual-branch feature extraction architecture that integrates multi-layer perceptrons with two-dimensional convolutional neural networks. This approach discerns device protocol structures, communication protocols, and network behaviors based on the extraction of numerical statistics and sequential features, respectively. An attention-gating mechanism accomplishes the dynamic weighted integration of diverse features. The simulation experiment's overall accuracy of the model on the IoT SENTINEL dataset attained 87.45%. In comparison to conventional Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN) models, it enhanced device recognition accuracy and improved the granularity of device recognition. The model possesses significant edge deployment potential as a reference for IoT device identification.