A novel method for derivation of habitat scene of some important wetland ecosystem in China

Abstract. A new approach was introduced for derivation of the typical wetland ecological habitat scene (WEHS) in China. The spatial envelope algorithm with the optimal configuration of multi-resolution filters in the spatial-frequency domain was proposed to extract the global semantic features of the WEHS. The amount of overlaps of scene features could be reduced when passing through such filter channels. The feature were then normalized and centralized. The principal component analysis algorithm was employed to project the preprocessed feature to the low-dimension subspace to extract the principal components. The multinomial logistic regression algorithm was carried out in such subspace. It generates generated 3-dimensional probability simplexes for each observation. The derivation of WEHS was specified according to the maximal probability. Finally, the unknown origin of the WEHS could be determined according to the value of the maximum similarity to the probability simplexes of the instances.