The SVD classification algorithm introduced in this paper is designed for remote sensing image processing. It intends to classify an image by simulating the procedure of human visual distinguishing spots. This method could avoid some defeats in normal classifications

decrease the mixture appeared at the edge or the juncture of classes

smooth the homogeneous spots and conserve the small isolate spots. The result of this new method presents advantages of regular spots

clear boundaries and high accuracy. It is obviously better than the results obtained by normal classifications and it is of great value in practical application.