|
In the second half of the 1990s, MAD (Multivariate Alteration Detection) method designed for detecting differences in multidimensional datasets was published. It was developed specifically for change detection performed on the basis of multispectral and hyperspectral images collected at different times. Within the framework of the European program SATChMo/Geoland2, the Space Research Centre of Polish Academy of Sciences has recognized the possibility of applying the IR-MAD transformation to detect changes of the main form of land cover on high resolution satellite images. Presented work was performed on the basis of a pair of KOMPSAT-2 images presenting area of Northern Spain in 2008 and 2010. The proposed algorithm analyses the results of the IR-MAD transformation and also additional information about vegetation cover and texture. Transformation of IR-MAD is performed on the basis of four spectral channels; B, G, R and IR with a resolution of 4 m, of the images from the first and second term. Additional information concerning a possibility of appearing changes in vegetation are derived on the basis of NDVI index and texture layer produced by Sigma filters of panchromatic channel of 1m resolution. An assumption was made for classified changes of land cover, which are independent of natural phenological cycles. First, places of changes are located and next information of their nature (direction of changes) is obtained.
|