A comparison of different land-use classification techniques for accurate monitoring of degraded coal-mining areas

Classification of different land features with similar spectral response is an enigmatical task for pixel-based classifiers, as most of these algorithms rely only on the spectral information of the satellite data. This study evaluated the performance of six major pixel-based land-use classification techniques (both common and advanced) for accurate classification of the heterogeneous land-use pattern of Jharia coalfield, India. WorldView-2 satellite data was used in the present study. The land-use classification results revealed that Maximum Likelihood classifier algorithm performed best out of the four common algorithms with an overall accuracy of about 84%. The advanced classifiers used in the study were Neural-Net and Support Vector Machine both of which gave excellent results with an overall accuracy of 91% and 95%, respectively. It was observed that use of very high-resolution data is not sufficient for obtaining high classification accuracy, selection of an appropriate classification algorithm is equally important to get better classification results. Advanced classifiers gave higher accuracy with minimal errors, hence, for critical planning and monitoring tasks these classifiers should be preferred.