부산대학교 도서관

In the field of hand-gesture recognition, the method of recognition by using single-class radar images has the problems of insufficient expression of features and poor accuracy. In order to solve these problems, a gesture recognition method based on radar multi-domain representation is proposed. Specifically, pulse compression and Capon beamforming algorithms are used to process radar signals to obtain two representations in different domains. The multi-channel parallel convolutional neural network is used to extract independent features from two types of radar images and optimize the features. Finally, feature fusion is performed to make the expression of features more sufficient. The experimental results show that the recognition rate of the method is at least improved by 3.2% compared with the method using single-class features. This method has comprehensive advantages in recognition accuracy and convergence speed.