A Multi-Head Attention-Based Method for Internet of Things Device Identification

With the gradual increase in the number of devices in the Internet of Things (IoT), accurate identification of devices has become crucial. Existing methods primarily rely on communication protocols, behavior characteristics, and traffic fingerprints, often failing to fully exploit the temporal characteristics of data traffic. This study proposes an IoT device identification method based on Multi-Head Attention. By extracting protocol features and packet features from traffic data, the method utilizes a multi-head attention mechanism to capture temporal features of packets, thereby achieving effective device identification. Experimental results demonstrate that the proposed method achieves a comprehensive recognition accuracy of over 97% on the Aalto dataset.