Thesis: Towards automatic lithological mapping from remote sensing data using support vector machines
Remote sensing data can be effectively used as a means to build geological knowledge for poorly mapped terrains. In this study, the support vector machine (SVM) algorithm is applied to an automated comprehensive lithological mapping of a study area in northwestern India using Advanced Space-borne Thermal Emission and Reflection Radiometer (ASTER) imagery, together with ASTER-derived digital elevation mode (DEM) and aeromagnetic data. Spectral techniques were used to produce derivative datasets from the ASTER image to enhance lithological information. The DEM and aeromagnetic data were also processed to generate derivative datasets relevant to lithological discrimination. A regional-scale geological map of the study area was used to select training samples. A series of SVMs were tested using various combinations of input datasets selected from 64 datasets including the original 14 ASTER bands and 50 derivative datasets extracted from the ASTER, DEM and aeromagnetic data, in order to determine the optimal inputs that provide the highest classification accuracy. A combination of ASTER-derived independent components, principal components and band ratios, DEM-derived slope, curvature and roughness, and aeromagnetic-derived mean and variance of magnetic susceptibility provided the highest classification accuracy of 93.43% on independent test samples. The output geological map shows a high level of similarity to the available regional-scale geological map. The study illustrates that SVM can be used to produce quick and reliable geological maps for areas with scarce geological information.