A Lightweight Deep Residual Network for Recognizing Activities in Daily Living Using Channel State Information

The field of human activity recognition (HAR) focuses on predicting human motion and actions by analyzing data from various sensors. HAR tasks can benefit from vision-based and sensor-based approaches, which offer high data quality. However, it's essential to recognize that these methods have drawbacks, such as intrusiveness and privacy concerns. With the increasing prevalence of Wi-Fi devices, Wi-Fi technology has gained popularity for tracking everyday activities, particularly among senior individuals, making it an appealing choice in healthcare applications. Channel State Information (CSI), a crucial attribute of Wi-Fi signals, can be leveraged to recognize different human actions. This study introduces CSI-ResNeXt, a lightweight deep residual network designed to classify human actions using CSI data from daily living scenarios. A series of tests were conducted to evaluate the network's performance, using a standard CSI-HA dataset. The results of the experiments demonstrated that the proposed lightweight deep residual network achieved an impressive accuracy of 99.17% and an F1-score of 99.10%, surpassing the performance of previous deep learning models.