Land Change Detection in Bitemporal and Multispectral Images Using Multivariate Alteration Detection and Statistical Thresholding Methods

In this paper, we shall describe a statistical approach for land change detection and extraction based on multivariate alteration detection (MAD) transformation combined with three thresholding methods. Unlike the most other multivariate change detection techniques, such as principal component analysis (PCA), the MAD analysis is invariant to linear and affine transformations of the input data. Consequently, it is insensitive to linear differences in atmospheric conditions or sensor calibrations of multitemporal acquisitions. Statistical thresholding methods which are based on 1) an Ad-hoc thresholding test, 2) Chi-2 thresholding test and, 3) ChiMerge thresholding test are applied to extract changes detected by means of MAD transformation. A case study with SPOT-HRV bitemporal and multispectral images before and after a flood event occurred in November 2000 over Gloucester (UK) region shows the efficiency and robustness of MAD/ChiMerge change detection scheme compared to MAD/Ad-hoc and MAD/Chi-2 schemes. A comparison of extracted change objects is also carried out according to the ground truth of the study zone.