Feature Extraction Method for DCP HRRP-Based Radar Target Recognition via $m-\chi$ Decomposition and Sparsity-Preserving Discriminant Correlation Analysis

Dual-circular polarimetric (DCP) high-resolution range profile (HRRP) provides similar target information to the full polarimetric HRRP and has the same data volume as the dual-linear polarimetric HRRP, thus it is significant to investigate the capability of DCP HRRP for target recognition. In this paper, a novel feature extraction method is proposed for the DCP HRRP-based target recognition. First, due to the good capability of characterizing the target polarimetric structure, the $m - \chi $ decomposition is exploited to obtain three scattering components of targets corresponding to the odd-bounce, even-bounce and randomly polarized scattering mechanisms along the radar line-of-sight (LOS), respectively. However, these three scattering components are infeasible to be directly utilized for target recognition due to the high dimensionality and redundancy. Considering this, a novel feature fusion method named as sparsity-preserving discriminant correlation analysis (SPDCA) is proposed to fuse the three scattering components. The SPDCA method can obtain the low-dimensional projection feature with good class separability by preserving both the global structure and local sparsity property of the original data. Besides, the redundancy of the three scattering components is eliminated by the SPDCA method. The results of experiments conducted on the simulated data of 10 civilian vehicles and real data of 3 military vehicles demonstrate the effectiveness and robustness of the proposed feature extraction method.