Radar HRRP recognition based on supervised exponential sparsity preserving projection with small training data size

Abstract

The echo signals from ships and sea clutter are coherently accumulated. Therefore, it is difficult to capture and distinguish the features within the signals. In addition, due to poor measurement conditions, the radar system can only collect data from a limited number of non-cooperative ships. In this article, a method termed supervised exponential sparsity preserving projection (E-MMC-SPP) is proposed for recognizing ship classes based on high-resolution range profile (HRRP). The method consists of three parts: First, to extract richer features from sea clutter, a maximum margin criterion sparse reconstructive relationship is constructed, which maximally preserves the sparse reconstruction of data and enhances class separability. Second, matrix exponential is utilized to ensure the positive definiteness of the coefficient matrices, thereby addressing the small-sample-size (SSS) problem. Finally, an efficient numerical method is presented for solving the corresponding large-scale matrix exponential eigenvalue problem. Experimental results on measured radar data demonstrate that the proposed method effectively reduces feature dimensionality and enhances target recognition performance with limited training data.