An improved SVM algorithm for high spatial resolution remote sensing image classification

Support vector machine ( SVM) algorithm has been widely used for remote sensing image classification . For high spatial resolution image classification , traditional SVM algorithm usually leads to low efficiency due to large quantities of high dimensional sample data .This paper presents a simple improved SVM algorithm with the purpose of improving both efficiency and accuracy of classification models .The algorithm first uses PCA to reduce the dimension of sample features .The grid -based method is used to search for optimal parameters for SVM classification of PCA-based samples .Then new range around the PCA -optimal parameters is set up and used for optimal parameter search based on the original sample data .Finally, SVM with the optimal parameters is used to train the original sample data and classify the image .The new algorithm was evaluated by two classification experiments based on WorldView 2 images including urban land cover land use classification and urban tree classification .Compared with the traditional SVM and SVM merely based on PCA data , the results show that the improved SVM algorithm could quickly and efficiently find the optimum parameters of the SVM classifier and achieves higher classification accuracy .