A method to improve the generalization ability of HRRP recognition model-Deep Adaptation Networks

Abstract

In order to improve the generalization ability of the deep neural networks, this paper used the Deep Adaptation Networks (DAN) to reduce the differences between domains, and applied it to the radar High Resolution Range Profile (HRRP) target recognition. In this paper, the one-dimensional convolutional neural networks (CNN) were used to extract the features of HRRP; the mixed kernel MMD replaced the Multi-Kernel MMD in DAN. And 2 adaptation layers were added to improve performance of recognition model. The HRRP target with noise was recognized in time and frequency domain respectively. The experimental results showed that, this method was better than the traditional transfer learning method or DAN. When the noise was small, this method improved the recognition accuracy to 90%; when the noise was large, it also improved the recognition performance. This method greatly enhanced the generalization ability of HRRP recognition network model and improved the recognition performance of the model in complex environment.