Secure Indoor Localization Against Adversarial Attacks Using DCGAN

Abstract

The vulnerability of deep learning-based indoor Wi-Fi fingerprint localization methods to adversarial attacks significantly reduces localization performance. To overcome this challenge, we propose a defense strategy employing a deep convolutional generative adversarial network (DCGAN) to enhance the security of channel state information (CSI)-based localization methods while maintaining accuracy. Our approach eliminates adversarial perturbations before the adversarial samples are fed into the deep learning model for localization. The localization performance of the proposed DCGAN is evaluated through experiments conducted with commodity Wi-Fi devices in representative indoor environments. Experimental results demonstrate that the DCGAN model effectively mitigates adversarial interference while maintaining excellent localization accuracy under two white-box attacks and one black-box attack.