基于FAST网络的毫米波雷达端到端手势识别OA

In view of the complexities,inefficiencies,and low accuracy of current millimeter-wave radar gesture recognition methods,this paper proposes an FAST(Fourier-Attention-Swin Transformer)network model.Firstly,a complex-valued linear layer is utilized to construct a Fourier network,and the weights of the Fourier network are initialized with discrete Fourier transform values.Range-Doppler features are obtained after radar' raw data passing through the Fourier network.Secondly,the ECA(efficient channel attention)module is introduced to calculate frame-channel attention weights,enhancing the capability to extract gesture features.Finally,the Swin Transformer is employed to improve computational efficiency and recognition accuracy while expanding the receptive field,and a loss function is used for backpropagation and iterative updates of the model parameters.Experimental results demonstrate that the proposed FAST-based end-to-end gesture recognition algorithm using millimeter-wave radar achieves an accuracy rate of 96.46%,showcasing advanced performance in comparison with the other mainstream algorithms.This study offers a more streamlined and efficient solution for the application of millimeter-wave radar gesture recognition in smart homes and mobile devices.