Detection of remote sensing image alteration based on canonical correlation analysis

Because of the effect of correlativity between channels, it is more difficult to detect the change of multi-channel remote sense images than to defect the change of single-channel images. Therefore effective concentration of the changing information on each channel is required to form difference images between different time phases so as to facilitate the analysis and interpretation of changing information. The problem is that it is difficult to concentrate multi-channel changing informatious and eliminate the effect of correlativity between multi-channel remote sense tmages. With regard to this problem, the canonical correlation analysis is introduced in multivariate statistical analysis. The authors take the multi-channel remote sense images in two time phases as two sets of multivariate random variables and adopt multivariate alteration detection to reorganize all difference information or changing information in spectral passages and distribute them in a group of uncorrelated variables. Thus the adverse effect of the correlativity between channels on the multivariate alteration detection is minimized and the problem of difference image structure is solved preliminarily.