Integration of Spectral Indices, Digital Elevation Data and Support Vector Machines for Land Use Classification in Hilly Areas

Abstract

In this work, to effectively improve land use classification accuracy in hilly areas, a new method by integrating spectral indices, digital elevation data and support vector machines (SVM), has been put forward. Firstly, the freely available Landsat ETM+ and ASTER GDEM data of the study area were downloaded and geo-referenced. Secondly, to reduce the topographic effects as well as to enhance the spectral discrepancies among different land use types, images of several widely used thematic-oriented spectral indices were derived and stacked together with the image of ASTER GDEM as input. Thirdly, the SVM, a classifier requiring no assumption of the underlying data distribution and working well even with small number of training samples, was applied to classify the input image. Finally, results from the method proposed were compared with conventional Maximum Likelihood Classification (MLC). The findings suggested that the new method performed better than the traditional MLC.