Radar HRRP Recognition using Attentional CNN with Multi-resolution Spectrograms

A novel high-resolution range profile (HRRP) target recognition method based on attentional convolutional neural network (CNN) model with multi-resolution spectrograms is proposed. Different from time domain HRRP, the multi-resolution spectrograms record the amplitude feature and characterize the phase information among range cells of the target at different time-frequency resolutions. The attentional CNN consists of several feature extractors and an attention mechanism to softly weight features extracted from multi-resolution spectrograms. Experimental results on measured HRRP data demonstrate the effectiveness of the proposed method, and diagnostically visualize the importance of the spectrogram at different resolutions.