Performance of image classification on hyperspectral imagery for lithological mapping

Abstract

SVM and SAM classifiers for the lithological mapping using Hyperion data in parts of Gadag schist belt of western Dharwar craton, Karnataka, India were used. The main objective of the present study is to assess and compare the potential use of Hyperion data set for lithological mapping. Accuracy assessment of the derived thematic maps was based on the analysis of the confusion matrix statistics computed for each classification map. For consistency, the same set of validation points were used in evaluating the accuracy of the lithological thematic maps produced. On the basis of the accuracy assessment results, it appears that SVM generally outperformed the SAM classifier in both OA accuracy and individual classes’ accuracies. OA accuracy and Kc for SVM is 96.93% and 0.9655, whereas for SAM it is 74.02% and 0.7085 respectively. SVM classification is the best in describing the spatial distribution and the cover density of each lithology, as was also indicated from the statistics of the individual class results. The individual class accuracy were also analyzed for the SVM and the result show that PA ranges from 87% to 100% and UA ranges from 91% to 100%, whereas for SAM ranges from 15% to 95%, and from 31% to 100% respectively. The SVM method could effectively classify and improve on the existing geological map for the Gadag schist belt (GSB) using hyperspectral data. The results could be validated through field visits. Therefore, it is concluded that hyperspectral remote sensing data can be efficiently used to improve existing maps, especially in areas where same rock types show variable degree of alteration over smaller spatial scales.