The "Soft" Satellite Classification Based on Dempster-Shafer of Evidence

Remote sensing classification is the core of converting satellite image to useful geographic information. Many methods have been proposed for improving classification accuracy, however, the results are always dissatisfied. The reason is that there is serious spectral overlay phenomenon between classes which decrease the classification accuracy. This paper introduced an evidential reasoning "soft" classification method which is built on the Dempster-Shafer of evidence,and we compared the classification accuracies between our method and traditional MLC method in Zhalong NNR. The results showed an improved classification results compared with MLC,leading to better discrimination, the overall classification increased significantly.