Multivariate Change Detection Based on Canonical Transformation
基于典型相关分析的多元变化的检测

The change detection is one of the important topics in multi_temporal remotely sensed data. The present paper introduces a method for multivariate change detection, which is based on the canonical correlation analysis and the orthogonal transformation. Moreover, an experiment with NOAA/AVHRR data is presented.Differing from traditional multivariate change detection schemes such as the principal component analysis (PCA), this method takes two multivariate or multi_spectral satellite images as a whole set; each image set (of both) covers the same geographic locations and is typically acquired at different times. Then the two_date image sets are transformed into one set of new random multivariate by using the canonical transformation. By doing so the correlation between the spectral bands in the same image and in the two_date images are removed so that the actual changes in all bands can be simultaneously and accurately detected. This method has been tested for inundation detection of Poyang Lake of China during the summer 1998 flood along Chang Jiang. The results were very promising. The method has a great potential for automatic change detection by using the multi_sensor and multi_temporal remotely sensed data.