关键词: 偏最小二乘, 最大期望, 高分辨率遥感影像, 变化检测, 多重相关性

Abstract:

In high resolution remote sensing images,there are lots of data redundancy and the small scale change threshold is difficult to accurately determine.To solve that problem,a high resolution image change detection algorithm based on partial least squares (PLS) and local maximum expectation (EM) is proposed.First,the remote sensing images of the multi band of the front and back phase are two sets of multivariate variables.Through multivariate statistical analysis of PLS transformation model,effective components are extracted and differential images are constructed by difference method.Then,the regions with relatively large local variations are used to propose the threshold of change detection for different images by EM algorithm.Finally,using the threshold determined by the local image,the whole difference image is divided into two categories:change and non change.And the filtering operator is applied to deal with the detection results.The experiment of high resolution remote sensing image data shows that the method can extract the change information of surface cover better by using multi spectral image,and it has certain application value for dynamic monitoring of geographical conditions.

Key words: partial least squares, maximum expectation, high resolution remote sensing image, change detection, multiple correlation