Mapping grassland vegetation cover based on Support Vector Machine and association rules

Support Vector Machine (SVM) has been used to classify data and extensively explored in various fields. Instead of using original data as model inputs, we proposed here SVM modeling based on a nonlinear-mapping approach. Such a nonlinear data mapping for the SVM increased the hyperplane margin space, decreased the structural risk minimization (SRM), and thus improved the performance of SVMs in respect to image classification accuracy. The proposed approach was tested to classify vegetation cover for typical grassland in Northern China based on Landsat ETM+ data. The performance of SVMs with the nonlinear data mapping approach was evaluated against that without data mapping, and also compared to similar studies using different approaches as well. The results indicated that, in terms of image classification accuracy, the proposed method achieved the best result (82.7% with kappa =0.80).