Deep one-class classification induced authentication for security protection of wearable IoT application

The Internet of Things (IoT) has become one of the most popular directions in the field of network and communication. However, with the rapid development of IoT, the number of terminal devices shows a geometric order of magnitude growth. The sensor network composed of sensing terminal devices can obtain information in all aspects, but at the same time, malicious attackers can also obtain the information. In order to solve this issue, this paper designs a lightweight authentication for security protection in wearable IoT environment through edge-cloud architecture. The proposed lightweight authentication system adopts PPG signal as biometric feature, which can be easily obtained and hard to forge. The authentication is implemented by using deep one-class classification model which is trained by using a PPG signal library. In order to avoid the limitation of resources at edge nodes, the deep one-class classification models are deployed at the cloud server. The experiments show that the proposed authentication system can achieve a promising result.