Small Sample Target Recognition Based on Radar HRRP and Sparse Autoencoder

In this paper, a small sample target recognition method based on radar High Resolution Range Profiles (HRRP) is proposed by combining sparse auto encoder (SAE) and clustering, which effectively improves the recognition accuracy of small sample target. Firstly, a SAE network is supervised learning by using small sample HRRP training data. Then the trained network architecture parameters and training data deep features could be obtained. And the centers and average distances of deep features clustering also could be obtained by K-means clustering. Subsequently, the deep features and distances of each test sample to various centers could be obtained. Then, by setting the distance threshold, some test data will be grouped into the training data set to expand the training data set. Finally, after loop iterations, the network architecture and recognition results could be obtained by using the final training data set. The results show that the proposed method can improve the recognition accuracy of space targets with high noise and small samples, which can provide a reference for the application of space target recognition based on radar electromagnetic scattering characteristics.