基于MAP和RF的无监督SAR图像变化检测

Unsupervised Approach Based on MAP and RF to Change Detection in Multitemporal SAR Images

基于形态学属性断面(MAP)和随机森林(RF)分类器,提出了无监督合成孔径雷达(SAR)图像变化检测方法.首先,利用MAP算法提取差异图像的几何结构特征,构造深入描述图像结构化信息的特征向量空间;然后,在结合阈值法和偏移因子自动选取训练样本的基础上,用RF分类器在多维特征空间中对图像进行变化与否的判别;最后,利用数学形态学方法对虚警进行滤除.实验结果表明,与传统的基于阈值的变化检测方法相比,该方法不仅能很好地检测出变化区域,而且具有更高的检测精度.

Based on the morphological attribute profile （MAP） and random forest （RF）, an unsuperviesed change detection approach for SAR images was proposed. Firstly, the MAP algorithm was employed to extract the geometric feature of the difference image and a feature vector space was constructed to describe the image inherent structure. Secondly, based on automatic selection of training samples by the combination of thresholding method as well as the offsets, the RF was employed to distinguish changed from unchanged pixels in multidimensional feature vector space. Finally, the mathematic morphology method was used to filter false alarm. Experimental results show that the proposed method can extract the changed area effectively and achieve a better performance than the classical change detection methods based on thresholding.