There are two ways to improve the performance of land cover classification with remote sensing images.The first way is to apply new data source including GIS data and normalized difference vegetation index(NDVI) to multi-source information fusion.The second one is to use methods with higher accuracy.Support vector machines(SVM) overcome the defects of maximum-likelihood and neural networks classifiers.SVMs are suitable to process complex data of high dimension and small number of training data.In this paper

selection of SVM models including kernel functions and multi-class methods is studied in order to improve the accuracy of multi-source remote sensing images classification.Experimental results show that the SVMs have higher accuracy than other traditional classifiers for the classification of multi-source remote sensing data.The SVM with a RBF kernel function and One-against-one multi-class method is the best classifier in this study.SVM methods could greatly improve the multi-source land cover classification.