mohammad mavadati

This work applies support vector machine (SVM) algorithms in two versions of singular and general SVM classifiers to map hydrothermal alteration zones in the northwestern part of the Kerman Cenozoic Magmatic Arc (KCMA). Three visible bands and six SWIR bands of ASTER images were applied as inputs for SVM classifiers. The develosped algorithms were able to classify ASTER images into hydrothermal alteration or non-hydrothermal alteration classes. In singular SVM, nine classifiers were able to vote individually for every pixel in the image. Then, they were combined through integration rules to present a final decision about every pixel. The general SVM classifier integrated nine ASTER bands at the signal level to produce the final decision. The classification error rate showed that the general Gaussian RBF kernel-based SVM classifier had higher accuracy for the classification of hydrothermal alteration zones. The SVM results were then compared with other classified images based on band ratio and SAM methods. The main problem associated with these methods was that vegetation covering was highlighted as alteration zones while the SVM algorithm could solve this issue. Also, the verification of results, based on field and laboratory investigations, showed the SVM method to produce a more accurate map of alteration than that obtained from the band ratio and SAM.