Wi-SSR: Wi-Fi-Based Lightweight High-Resolution Model for Human Activity Recognition

Abstract

In recent years, human activity recognition (HAR) based on Wi-Fi channel state information (CSI) has received widespread attention due to its non-intrusive and privacy-preserving nature. However, many CSI activity recognition models based on traditional methods and deep learning face two major challenges: first, most studies rely on commercial Wi-Fi network cards, which usually have only three RF ports, resulting in limited spatiotemporal resolution of the acquired CSI; second, some of the studies require complex CSI processing, which increases the network parameters, significantly lengthens the recognition time and raises the deployment costs. To this end, this study develops a lightweight high-resolution recognition model Wi-SSR based on Wi-Fi. To improve the spatiotemporal resolution of CSI, we introduce array antennas and solve the problem of coherent signals that are difficult to distinguish by communication algorithms. The lightweight CSI processing strategy proposed by Wi-SSR is able to efficiently extract the main relevant features while compressing the model size. We combine 3-D convolution with a convolutional block attention module (CBAM) to extract activity-related information from CSI and employ knowledge distillation to migrate the features learned from this model to a simple model. Extensive experimental results show that our system outperforms other deep learning models in terms of efficiency, with recognition accuracy up to 98.6% on six different types of human activities.