RESUMEN
Aiming at the problem that a convenient multivariate statistical model is in general not available for the multi-spectrum feature of land use/cover (LUC) class in remote sensing (RS) image, because the class is made of multiple covered species, a spatial-distance analysis approach of multi-spectrum feature distribution for RS image LUC is present, with the mean vector of samples as LUC class center, with max-min clustering algorithm forming the class multi-clustering-centers, the spatial-distances from the class center to these multi-clustering-centers were calculated. With the distance as abscissa and the percentage of the clustering-center pixels to the whole sample pixels as ordinate, the intra- and inter-classes distance distribution charts were constructed to analyze the multi-spectrum feature distribution of RS image LUC. The results of these samples classification tally with the conclusions of spatial distance analysis, indicating that this approach is feasible. In this approach the multi-dimensional spectrum information is turned into one dimensional distance information, the spatial-distance calculation and clustering threshold confirmation are realized easily, and the multi-spectrum feature of LUC class is clear, so it is a better approach to solving the multivariate distributing problem of multi-spectrum feature.