Multi-classifier fusion for land use land cover mapping in Jharia coal field

Abstract

Open cast mining process destroys landscapes, forests and wildlife habitats due to clearance of trees, plants, and topsoil. This results in ecological degradation such as soil erosion, destruction of agricultural land, water pollution, etc. Therefore mapping and monitoring of landscapes for temporal detection is extremely essential. A study of application of data fusion techniques for mapping of land use cover has been carried out for a mining coalfield in order to ascertain potential impact of mining activities on flora and fauna. The study has incorporated pixel level Image fusion process as well as decision level fusion procedures. High Pass Filter (HPF) method based on pixel level fusion is applied to fuse the multispectral (MS) and panchromatic (PAN) image of Landsat ETM+ data. The fused MS image is classified into seven land use categories viz. agricultural land or open shrub, water bodies, dense vegetation, built-up land, barren land, mining area and overburden dump area using different classifiers. The selected classifiers are maximum likelihood, mahalanobis distance and support vector machine. The Naive Bayes classifier fusion rule on abstract level is applied to integrate the decisions from each classifier. The result of accuracy assessment shows an little increment of 0.41% and 0.73% in both classification accuracy as well as kappa coefficient in comparison to single classifier with the highest accuracy.