Abstract: with the rapid development of the Internet of things (IoT), IoT security issues have received widespread attention. The hardware and software features of IoT devices make them extremely vulnerable to all types of attacks. Anomaly detection of IoT devices has become a hot spot in recent years. The traditional protection methods based on intrusion detection and traffic analysis can not adapt to the hardware and software environment of IoT devices. In order to solve this problem, an anomaly detection scheme based on chip radiation is proposed. By using the electromagnetic wave signals of IoT devices radiating outwards during operation as detection basis, the original signals are extracted and selected by genetic algorithm and approximate entropy. Finally, the signal of normal behavior radiation is trained using a one-class support vector machine algorithm. The program has non-invasive features, without the need for any transformation of the original system hardware and software, applying to the existing IoT devices. The final experimental results show that compared with other commonly used anomaly detection schemes, this scheme can detect the abnormal behavior of IoT devices more effectively, with higher accuracy and lower false alarm rate.

Key words: anomaly detection, IoT device, radio emission, one-class SVM, clustering