Research on HRRP target recognition based on one-dimensional stack convolutional autoencoder

Abstract

Extracting key features from the radar high resolution range profile (HRRP) determines the accuracy and reliability of radar target recognition. Aiming at the problem of feature extraction and recognition in HRRP target recognition, a one-dimensional stack convolutional autoencoder (1D-sCAE) recognition method is proposed. Firstly, one-dimensional convolutional autoencoder is constructed to extract the deep features of one-dimensional HRRP signals by unsupervised learning. Then, multiple one-dimensional convolutional autoencoders are stacked to construct 1D-sCAE, and HRRP is classified by fine-tuning the network with label data. At the same time, for the overfitting problem of 1D-sCAE, dropout technology is added to optimize and improve the generalization performance. Through the simulated HRRP data of the target in the middle part of the trajectory, the experiment shows that the proposed algorithm has better feature extraction capabilities, higher recognition accuracy and stronger robustness than the typical deep neural networks.