Classification Of Remote Sensing Image Based On Multi-layer Controlled Combining Classiiiers

Remote sensing image classification is one of the most efficient ways to extract landuse/cover information. However, there are many factors that may reduce the classificationaccuracy, such as same-object-different-images and different-objects-same-image. Moreover,constrained by a single classifier and targeted restirction, overall accuracy is difficult to meet thestandard of extracting land use/cover information or accuracy of some categoires fails to meet.This paper designs a multi-layer controlled combining classifiers algorithm to solve theseproblems. It predicts the number and combinations of single classifiers in combining classifiers,using of EPD diversity measure. Meanwhile, the remote sensing image can be reclassified basedon the weight vote method and AdaBoost.M2and we can define accuracy and stability ofcredibility coefficient of classification image. Through compairng credibility coeiffcient ofclassification image between the multi-layer controlled and Bayesian single controlledcombining classifiers, it is validated that EPD diversity measure can improve the eiffciency andoverall performance in combining classifiers by reducing the number of classifiers and dataredundancy. It proves that with better robustness and generalization, the multi-layer obtainshigher classiifcation accuracy and credibility coefficient than single-layer.