Hyperspectral Image Classification using Machine Learning

Abstract

Hyperspectral images are of high resolution so helpful in classifying the land surface for variety of applications like land use and land cover, lithological discrimination, differentiating different land surfaces, segmentation, etc. This work focused on the hyperspectral image classification at Lonar Crater situated at Buldhana district, Maharashtra. This methodology consists of sequence of operations viz. bad band removal, destriping, atmospheric correction, minimum noise fraction, n-D visualization and classification. 242 images of Hyperion sensor are used for classification. The land surface at and near Lonar crater is classified with more user accuracy on an average. The Spectral angle mapper technique is used showing the 59.23% accuracy in classifying with respect to ground truths from image. The results are improved with Support Vector Machine.