CNN Based on Multiscale Window Self-Attention Mechanism for Radar HRRP Target Recognition

In this paper, we propose a convolutional neural network (CNN) with multiscale window self-attention mechanism for radar high-resolution range profile (HRRP) target recognition task. Specifically, we take a one-dimensional CNN (1-D CNN) to extract shallow feature and fully excavate the rich local structural features of HRRP data. Then, we utilize a multiscale window convolutional self-attention mechanism to capture the regional difference of HRRP data. The proposed self-attention module divides the features obtained by CNN into equal width bands through sliding windows. Multi-level features of different regions can be obtained by the continuous expansion of the windows. It can improve the model's attention to target regions and suppress the influence of irrelevant noise. Experimental results show the superiority of the proposed method in HRRP recognition task.