HRRP Target Recognition Based On Soft-Boundary Deep SVDD With LSTM

Radar high-resolution range profile (HRRP) target recognition is an active part of radar target recognition (ATR). The current radar HRRP target data has the characteristics of less data volume. At the same time, support vector data description has limited ability to extract the deep features of the signal. To address this issue, we propose a soft-boundary Deep SVDD with LSTM (long short-term memory). The framework consists of an autoencoder, an LSTM neural network layer, and an SVDD hyper-sphere. The autoencoder generates the deep signal features, and the LSTM layer can extract the time-related features. The distance from the feature point to the center of the hyper-sphere is the classification judgment condition. The neural network parameters and the hyper-sphere are trained to be the optimal value. We carry out experiments on a dataset with a small volume. The result and the Received operation characteristic (ROC) curve show that the classifier has good performance. The area under ROC (AUC) value is close to 87%–94%.