Continuous Head Motion Recognition Method for Millimeter-Wave Radar

In this paper, a continuous head motion recognition method of millimeter wave radar is studied. Firstly, a combination of standard convolution and depthwise separable convolution is used to extract features, which reduces the number of parameters and the size of the model while ensuring the accuracy. Then the Coordinate Attention mechanism is combined to filter the features, so that the network pays more attention to the important features. The feature dimension is reduced by using max pooling instead of full connection layer, and the parameter number and computation cost are reduced. The LSTM network is then used to train the time series data. Finally, the proposed method is verified by the measured data, and its effectiveness is proved.