Seasonal detection of coal overburden dump regions in unsupervised manner using landsat 8 OLI/TIRS images at jharia coal fields

Abstract

Classification and monitoring of surface mining areas have various research challenges. Surface mining produces various land classes such as, quarry, dump, overburden dump, reclamation area, etc. In the past, various land classes of surface mining areas are detected by supervised and semi-supervised machine learning techniques. It has been found challenging to detect such land classes using only spectral responses from satellite images. Coal Mine Index (CMI) detects coal quarry and coal dump region as a single land class. These regions have distinct properties as minerals stayed open in such regions. Though coal overburden dump regions also have higher mineral content than various land classes, they show similar spectral characteristics with few bare soil classes in particular with river beds. Hence, it is found more challenging to detect coal overburden regions in an unsupervised manner using spectral information. In this paper, a K-Means clustering in hierarchical fashion has been proposed using CMI values as feature space to detect coal overburden dump regions in automated manner. Yet, this procedure detects coal overburden dump and river beds as a single class. The method is further extended to distinguish river bed regions from coal overburden regions exploiting their distinctive spectral characteristics. The proposed method has average precision and recall of [76.43%,62.75%], and [70.37%,65.63%] for coal mine, and overburden dump regions, respectively.