SVM and random forest classification of satellite image with NDVI as an additional attribute to the dataset

Abstract

Image Classification is a widely researched subject both in the academic circles as well as industry having varied use in wide ranging applications in a multidisciplinary area. These techniques are being used in the fields of robotics, military applications, telecom, Oil and gas exploration, agriculture, creation of mapping products for military/civil applications and also by various departments of the Government for assessment of environmental damage, land use monitoring, radiation monitoring, urban planning, growth regulation, soil evaluation, and crop yield assessment. This paper addresses the issue of use of SVM and Random Forest algorithms for the classification of the Alwar satellite image data set. The normalized difference vegetation index (NDVI) is an estimate of the photosynthetically absorbed radiation over the land surfaces. We use NDVI as an additional attribute in the satellite image dataset to study the impact on the classification accuracy by these two algorithms. The experiments were performed using Alwar satellite image dataset, which has been previously classified using various other classification algorithms. The prediction accuracy and the kappa coefficient achieved for training data and the classified image generated from the test dataset after use of both the classifier systems is finally compared.