A COMPARISON OF MULTI-CLASS SVM STRATEGIES AND KERNEL FUNCTIONS FOR LAND COVER CLASSIFICATION

Support Vector Machines (SVMs) are powerful machine learning algorithms originally proposed for solving linear and binary problems and, later, extended to perform non-linear and multi-class tasks. In remote sensing applications, SVMs have been widely applied to land cover classification. However, SVMs are highly sensitive to the choice of the kernel function and its parameters. These elements have a direct influence on the classification accuracy. The purpose of this study is to assess the performance of the SVM classifier when combined with distinct kernel functions and multi-class approaches for land cover classification. We carried out experiments using a multispectral image of a highly urbanized area. The experimental results demonstrated the efficiency of the SVM classifier with the radial basis function for land cover classification. In this study, the type of multi-class approach did not present a significant impact on the SVM performance when combined with this kernel function.