Multiple Classifiers Fusion For Remote Sensing Image Classification

In remote sensing image classification applications, the classification accuracy of different classifier is different. The difference is related with the statistical distribution characteristics of classification data, prior knowledge, the size of trained samples and structure of itself. Combination of multiple classifiers is a method which uses the complementary of the existing classification and finds an appropriate way through which we can use the advantage of different classifiers. Generally, we can get a better one than the classification results of a single classifier.This paper studies how we can use the shortest distance classifier, Mahalanobis classifier, Bayesian classifier, K-MEAN classifier, BP neural network classifier for image classification with Matlab. Then we use the voting method and Bayesian method to do remote sensing image classification and use the method of Confusion matrix to analysis the accuracy. The experimental result shows that the accuracy of combination of multiple classifiers is higher than a single classifier for remote sensing image classification.