WiFi-Based Lightweight Gesture Recognition for Coal Miners

With the advancement of smart mines, the need for gesture recognition for remote interaction between underground workers and machines has become crucial. However, traditional gesture recognition techniques require complex models that are very difficult to be deployed to the edge. To address this challenge, a gesture recognition method based on knowledge distillation is proposed in this study. First, the CSI ratio model is used to eliminate phase error and environmental noise, followed by the application of discrete wavelet transform to eliminate hardware noise interference. Then, the processed data is adaptively segmented using the principal component analysis and local anomaly factor algorithm to eliminate redundant static components. After that the processed CSI data is transformed into images using the relative position matrix method. Finally, knowledge distillation is employed to migrate knowledge from a teacher model to a student model, reducing the number of model parameters. Experiments conducted on the proposed method showed that it can achieve a recognition accuracy of 94.2% for hand gesture detection, which meets the requirement for gesture recognition in the mining industry.