Attack Evaluations of Deep Learning Empowered WiFi Sensing in IoT Systems

Abstract

Recently, the market has witnessed fabulous growths of WiFi sensing technologies to advance the deployment of Internet of Things (IoT) systems. Besides being an essential IoT enabler, WiFi devices with smart sensing techniques carry more helpful information to improve the performance of IoT systems. However, there is a lack of research efforts on systematically investigating the potential security threats of smart WiFi Sensing assisted IoT systems. In this paper, we systematically investigate the vulnerability of existing deep learning empowered WiFi sensing systems under False Data Injection (FDI) attacks. First, we analyze the attack vendors in three different IoT layers and consider violating the data integrity in three dimensions (i.e., time, space, and value). Second, we design three attack schemes to realize FDI attacks on smart WiFi sensing models. By measuring the performance of activity recognition on seven datasets under diverse WiFi sensing, experimental results demonstrate that the accuracy of activity recognition drastically decreases under our attacks.