Within the technical framework about "multi-level remote sensing alteration information separation and extraction" the complex remote sensing alteration information extraction is transformed to the analysis of the three targets i.e., background, interference and alteration of a remote sensing image on basis of diagnosis of the remote sensing image characteristics. The geometric structural analysis of spectral data space (also known as feature space) is applied to the research and spatial orientation of the clustering structure of the three targets, especially for the identification and division of interference factors. By means of M value estimation of end member of multi-band remote sensing data the background-abnormal subspace is determined. Then characteristic bands for the information extraction are chosen with regression skewness curve analysis and spectral reflectance characteristics of surface features.