An intelligent wavelet transform-based framework to detect subsurface fires with NOAA–AVHRR images

Abstract

Subsurface coal fires (in this article, termed as hotspots), responsible for atmospheric pollution, human fatalities and perilous land subsidence, pose a big threat to major coal-producing countries in the world. The majority of the research performed to date has focused on providing hotspot allocation information for a specific region of interest and most has explored quite expensive high-resolution Landsat Thematic Mapper (TM) satellite images for the same. This article aims to investigate the applicability of a wavelet transform-based model to detect subsurface fires (hotspots) with freely available National Oceanic and Atmospheric Administration/Advanced Very High Resolution Radiometer (NOAA–AVHRR) images and find out the possibility of extracting novel hotspot features by applying a wavelet transform-based analysis technique. The proposed feature vector consists of wavelet variance coefficients (WVCs) obtained from scale-by-scale decomposition of the AVHRR image variance and builds up a strong base for designing an accurate classification system. Furthermore, the support vector machine (SVM), an efficient machine learning tool, is applied to the proposed feature vector in order to develop a classification model. The demonstrated results successfully prove the effectiveness of the proposed framework as the classified images show a good correspondence with records of subsurface fires mapped by the Bharat Coking Coal Limited (BCCL), India. The effectiveness of the SVM method is also evaluated in comparison with the classical neural network-based approach.