SwinFMCW: A Joint Swin Transformer and LSTM Method for Gesture and Identity Recognition Using FMCW Radar

For the problem of data distortion and inadequate feature extraction in FMCW radar-based gesture and identity multitasking recognition under smart home, smart medical, and game interaction fields. First, user data are collected using FMCW radar, then, micro-Doppler features are constructed which are pre-processed using Bicubic interpolation combined with RobustScaler, and finally, a joint Swin Transformer and BiLSTM neural network is proposed for multi-task learning. The experiments show that the recognition accuracy of the proposed method in this paper is 95.7% and 94.8% on 8 gesture categories and 9 identity categories, which surpasses the existing methods and can effectively accomplish gesture and identity multi-task recognition.