A Rough Set Classifilcation Based Approach to Detect Hotspots in NOAA/AVHRR Images

India accounts for the greatest concentration of coal fires in world. Nearly half of the subsurface mine fires (hotspots) in Indian coalfields exist in Jharia (Jharkhand) region. Careful attention is required in this direction for mapping, monitoring and detecting these hotspots. Present paper utilizes the potential of operational satellite images to detect hotspots in Jharia region. Proposed algorithm consists of two steps: (1) marking potential hotspot pixels in NOAA/AVHRR image using different AVHRR channel statistics (i.e. average & variance), and (2) generating rules using the potential hotspots pixel information obtained in first step, in order to classify seen or unseen AVHRR images in hotspots and non-hotspots classes. Rough set theory is emerging as a new powerful tool for learning classification rules. In this paper, we propose a rough set based method to classify NOAA/A VHRR images of Jharia region in order to determine the spatial allocation of hotspots. Instead of applying all induced rules for classifying AVHRR images, only those generated rules take part in the classification process which meet the user specified criteria, thus simplifying the whole classification procedure. Proposed algorithm appears to detect hotspots successfully with throughout greater than 90% classification accuracy.