基于改进正则化方法的SAR图像区域特征提取

Region Feature Extraction Based on Improved Regularization Method in SAR Image

合成孔径雷达(SAR)图像上的各种噪声削弱了目标、阴影等感兴趣区域(region of interest,ROI)的细节特征,影响了后续的目标检测、分类和识别等应用。传统的正则化方法能够增强SAR图像的目标特征,但是运算量过大,实时性不好。提出一种改进的正则化方法,有效地提高了SAR图像区域特征提取的速度和精度。理论上证明,降质算子的优化可以使运算量由O(M3N3)降到O(MN),同时保留了区域特征增强的能力。利用MSTAR数据库中实测的SAR图像进行算法验证,实验结果表明该方法能够大幅度提高目标杂波比,有效抑制感兴趣区域内的噪声,从而更精确地把目标和阴影等区域从背景杂波中提取出来。

The noise existed in Synthetic Aperture Radar （SAR） image weakens the detailed features of region of interest （ROI） such as target and shadow. It also leads to the serious performance reduction of subsequent target detection, classification and recognition. The traditional regularization method could enhance target features in SAR image; however, the high computation complexity limits the real-time application of it. An improved regularization method is introduced, which increases both speed and precision of region feature extraction for SAR image significantly. It is theoretically proved that, by optimizing SAR projection operator, computation complexity could be reduced from O（M^3N^3） to O（MN） without ability losing of the region-based feature enhancement. MSTAR SAR image data is employed for algorithm experiment. The result shows that our method can increase target-to-clutter ratio significantly while restraining the noise in ROI, and then extract target and shadow from background clutters in SAR image more accurately.