Hierarchical Sequential Feature Extraction Network for Radar Target Recognition Based on HRRP

Radar high resolution range profile (HRRP) contains lots of discriminative information for radar automatic target recognition (RATR). The temporal dependence information in HRRP can provide the scatter distribution information at different times, which is particularly important for RATR and attracts intensive attention among researchers. Recently dynamic methods have been proposed to extract time sequential features for RATR and achieved promising performance. However, most of these time sequential features extraction methods only focus on temporal dependence information between HRRP range cells and pay little attention to the time sequential features of HRRP sequences. Aiming at this problem, a novel hierarchical time sequential feature extraction network is proposed in this paper. In the proposed method, we firstly transform one whole HRRP sample into several small HRRP sequences by the sequential information preprocessing (SIP) module, and then, we proposed the time sequential feature between HRRP range cells (TSFR) module and time sequential feature between HRRP sequences (TSFS) module to extract hierarchical sequential features. Compared with other time sequential feature extraction methods, the proposed TSFS module extracted time sequential features of HRRP sequences based on the information of temporal dependence between HRRP range cells, rather than from HRRP sequences directly. Besides, we also employed a feature fusion module to improve the stability of the model. In order to demonstrate the effectiveness of proposed method, experiments on an airplane electromagnetic calculation dataset were conducted and experimental results showed the superiority of it. At last, comparative experiments have been conducted to explore the contribution of each module in the hierarchical time sequential feature extraction network.