基于超像素分割与CRF的高分辨率遥感影像变化检测

针对目前高分辨率遥感影像变化检测算法对于光谱变化过敏感问题，本文提出了一种基于超像素分割与条件随机场（CRF）的遥感影像变化检测算法。首先采用空间约束的t混合模型驱动的分割模型，获得同质性超像素块，实现良好的边界附着性和亮度均匀性。然后计算分割得到的双时相影像块之间的特征差异性，获取变化幅度图像。最后利用模糊聚类算法（FCM）对变化幅度图像进行聚类，得到隶属度图像作为CRF一阶势，并利用光谱-空间相似度约束的函数构建CRF二阶势。试验结果表明，与现有方法相比，该方法检测精度可提高5%，错检率和漏检率可降低3%，能较好地应对输入图像的光谱变化，并保持变化检测结果的边缘细节。

Change detection based on superpixel segmentation and CRF for high-resolution remote sensing images

High-resolution remote sensing image change detection algorithms are too sensitive to spectral changes, a new algorithm based on super-pixel segmentation and improved conditional random field (CRF) is proposed in this paper to detect changes in remote sensing images. Firstly, the student's-t mixture model (SMM) with spatial constraints super-pixel segmentation model is used to generate homogeneous patches,which has characteristic of boundary adhesion and brightness uniformity. Then the algorithm obtains change amplitude image by calculating the feature difference between the segmented dual phase image patches. Finally, the change amplitude image is clustered by FCM, and clustering result is used as the first-order potential of CRF, and the spectral spatial similarity constrained function is used as the second-order potential of CRF. Experiments results show that the detection accuracy of the method proposed in this paper is improved by 5%, and the false detection rate and missed detection rate are reduced by 3%, this method can better handle the spectral changes of the input image and keep the edge details of the change detection results.