Abstract
Detecting disasters in underground mining, such as explosions and structuraldamage, has been a persistent challenge over the years. This problem iscompounded for first responders, who often have no clear information about theextent or nature of the damage within the mine. The poor-light or even totaldarkness inside the mines makes rescue efforts incredibly difficult, leading toa tragic loss of life. In this paper, we propose a novel instance segmentationmethod called DIS-Mine, specifically designed to identify disaster-affectedareas within underground mines under low-light or poor visibility conditions,aiding first responders in rescue efforts. DIS-Mine is capable of detectingobjects in images, even in complete darkness, by addressing challenges such ashigh noise, color distortions, and reduced contrast. The key innovations ofDIS-Mine are built upon four core components: i) Image brightness improvement,ii) Instance segmentation with SAM integration, iii) Mask R-CNN-basedsegmentation, and iv) Mask alignment with feature matching. On top of that, wehave collected real-world images from an experimental underground mine,introducing a new dataset named ImageMine, specifically gathered inlow-visibility conditions. This dataset serves to validate the performance ofDIS-Mine in realistic, challenging environments. Our comprehensive experimentson the ImageMine dataset, as well as on various other datasets demonstrate thatDIS-Mine achieves a superior F1 score of 86.0% and mIoU of 72.0%, outperformingstate-of-the-art instance segmentation methods, with at least 15x improvementand up to 80% higher precision in object detection.