面向SAR图像目标识别的鲁棒处理算法

Robust processing algorithm for SAR image target recognition

现有合成孔径雷达图像的目标识别方法通常要进行预处理,预处理对于识别率影响较大。但是,针对不同的合成孔径雷达目标图像,预处理算法的自适应性很难得到保证。将基于核的主成分分析与稀疏表示相结合,只需很少的观测数据就能得到高识别率的目标识别结果,节省了数据存储量和计算量。首先,阐述了压缩感知的基本理论;其次,提出了基于核主成分分析和稀疏表示的合成孔径雷达图像目标识别算法;最后,选取MSTAR数据库中的5类目标进行实验。仿真结果表明,在没有方位角预测的情况下,该算法仍能有效地识别目标,与其他识别算法相比,在同等噪声污染的图像下,具有较高的识别率。

With the existing synthetic aperture radar （SAR） image target recognition algorithm, image pre-processing has to be usually carried out. Preprocessing has a significant impact on the recognition rate. Howev-er, the adaptability of the preprocessing algorithm is difficult to be guaranteed. This paper proposes to apply the theory of kernel principal component analysis （KPCA） and sparse representation to the image to he recognited, thus achieving a target recognition result possessing a high recognition rate with only a few observation data and saving the data storage and computation. This paper describes the basic theory of compressed sensing first, and proposes an SAR image target recognition algorithm based on KPCA and sparse representation. An experiment is carried out with five kinds of SAR targets in the MSTAR database. The simulation results show that this pro-posed algorithm is still able to recognize the target effectively without prediction of the attitude angle. Compared with other recognition algorithms, it has a higher recognition rate to the image under the noise pollution on an e-qual basis.