Classification of ASTER image using SVM and local spatial statistics Gi

In this paper, the SVM classifier with RBF kernel function was utilized to tackle the classification of ASTER remote sensing image. Instead of the original image, the image of Gi, which is a statistics describing the local spatial structure, is inputted to the SVM classifier to get the final classification result. The classifying process includes a "probing stage" and a "classifying stage". The objective of the "probing stage" is to find an optimal lag value of Gi; and in the "classifying stage", the Gi image with the optimal lag is classified by the SVM classifier. The experimental result shows that Gi images with appropriate lag values can be used to distinguish land covering features with similar spectral characteristics and different local spatial structures and, as a result, to improve the overall classification accuracy.