Evolutionary support vector machine and its application in remote sensing imagery classification

The regularization parameter and the kernel parameters greatly affect the performance of support vector machines (SVM) models. This paper proposes an evolutionary algorithm (EA) to automatically determine the optimal parameters of SVM with the better classification accuracy and generalization ability simultaneously. The proposed ESVM model, called evolutionary SVM or ESVM, was applied to a Land-cover classification experiment in a 840×840 pixels Landsat-7 Enhanced Thematic Mapper plus (ETM+) high-resolution image of Wuhan in Hubei province of China compared with the conventional SVM model. Experimental results show that the use of EA for finding the optimal parameters results mainly in improvements in overall accuracy and generalization ability in comparison with conventional SVM. It is observed that classification accuracy of up to 91% is achievable for Landsat data produced by ESVM.