基于目标相干性表征差异的多波段SAR相干变化检测方法

相干变化检测(Coherent Change Detection, CCD)利用变化前后SAR图像间的相位相干性检测场景中发生的微小变化。传统的CCD方法由于对目标探测尺度单一，难以区分场景中目标变化区域与低相干干扰区域。多波段SAR对目标进行多尺度探测，依据电磁波对目标的穿透特性、目标的结构特性以及目标发生的变化尺度形成不同的相干性表征。该文据此提出一种多波段CCD方法。该方法先分别获取各个波段的相干变化差异图，然后依据目标的多波段相干性表征使用改进的期望最大化(Expectation-Maximization, EM)算法对场景分类，接下来根据少量监督样本确定目标变化类别，最后用Dempster-Shafer (DS)证据理论处理，获取多波段融合相干变化差异图。该结果可有效排除各个单波段存在的低相干干扰，达到降低虚警概率的目的。该文采用变化前后的L波段与P波段重轨SAR数据进行方法验证，实验结果与指标参数证明了该方法的有效性与正确性。 

Multi-band SAR Coherent Change Detection Method Based on Coherent Representation Differences of Targets

Coherent Change Detection (CCD) detects micro changes in a scene using phase coherence of SAR images before and after a change. It is difficult for conventional CCD method to distinguish low coherence interference region from objective change region because of a single detection scale. Multi-band SAR target detection in a multiscale way develops variable coherent representation according to the diversity of electromagnetic wave penetration, target structure, and change magnitude. In this paper, a multi-band CCD method is proposed. Firstly, the CCD images of every band were acquired; secondly, the detected scene was classified on the basis of coherent representation of targets in each single band using improved Expectation-Maximization (EM) algorithm; thirdly, the objective change class in each single band was selected by a few supervised samples; lastly, multi-band fusion CCD image was acquired by the use of Dempster-Shafer (DS) evidence theory. This multi-band CCD result can eliminate low coherent interference in each single band and decrease false alarm probability. The method is validated by L- and P-band repeat-pass SAR images acquired before and after change. Results and index parameters demonstrate the validity and correctness of the proposed method.