基于超像素融合的遥感影像变化检测方法

Remote sensing image change detection method based on superpixel fusion

以基于像素变化检测方法为基础,提出一种基于超像素融合的遥感影像变化检测方法。该方法主要包括以下步骤:首先对原始遥感影像进行两次超像素分割,生成像素级、超像素Ⅰ级和超像素Ⅱ级三个不同尺度的分割图,然后分别对每个分割图进行变化矢量分析,得到三个尺度的差分影像。分别对多个差分影像进行模糊C均值聚类,得到三个尺度变化与不变的隶属度,最后将三组隶属度作为证据,运用Dempster-Shafer证据理论进行融合,得到最终变化检测图。该算法能够结合像素级和超像素级影像的优势,很大程度上解决像素级方法效果不佳的问题。一组遥感影像试验结果表明所提方法是可行、有效的。

Based on the pixel-level change detection method, a remote sensing image change detection method based on superpixel-level fusion is proposed. The method mainly includes the following steps: firstly, the original remote sensing image is segmented twice to generate pixel-level, superpixel-level Ⅰ and superpixel-level Ⅱ three different scale segmentation map. Then each segmentation image is analyzed by change vector to obtain three scale difference images. Fuzzy c-means clus tering is carried out on multiple difference images to obtain three scale change and invariant membership degrees. Finally, the three groups of membership degrees are used as evidence and fused by using Dempster Shafer evidence theory to obtain the final change detection map. The algorithm can combine the advantages of pixel-level and superpixel-level images, and solve the problem of poor effect of pixel-level methods to a great extent. The experimental results of a group of remote sensing images show that the proposed method is feasible and effective.