M Transformation Used in Multivariate Change Detection of NOAA/AVHRR Data

The fundamental theory of orthogonal transformation based on canonical correlation analysis is applied to multivariate change detection with NOAA/AVHRR data.With two multivariate satellite images covering the same geographic area acquired at different points in time as a random whole sample,the strategy transforms two sets of random variables into one set of new random multivariate through the so-called canonical transformation introduced in the paper. It is tested in the middle reaches of Yangzi river where a great and heavy flood was caused during the period of time from July to August in 1998.The purpose of this experiment is to investigate what kinds of changes happened in this area from September,1996 to August,1998.At the same time,the experiments of the other methods for change detection have been conducted,comparing the result of the method presented in this paper.By comparison the experimental result shows the fact that the method based on canonical correlation analysis is exactly creditable and effective not only for clearly detecting water changes caused by flood,but also other changes such as vegetation,cloud,and so on.But with other methods,such as simple image difference and principal component analysis,it is difficult to distinguish the changes between water and vegetation.