Image Segmentation Using Deep Learning for Mining Road Width and Slope Alerting in Berau, East Kalimantan, Indonesia

Abstract

Mining roads are connecting infrastructure which has essential role in mining operations. The road conditions including road path, width and slope could change in different directions at any time. As a result, there might be accidents or fatality because of these changes. Therefore, mining roads need to be monitored regularly for maintenance and reconstruction to support operation safety. However, there are some challenges in monitoring mining roads, e.g. difficulties in interpreting the road object on imagery because of its materials. An improved interpretation technique to extract mining roads on imagery is needed for effective road identification. DeepLabV3 algorithm was applied in this research to conduct semantic segmentation for road interpretation on imagery. The output of image segmentation was used for the measurements and alerting of road width and slope in spatial approach. With the accuracy of 76.5%, DeepLabV3 could support mining road condition monitoring through image segmentation in acceptable quality. The integration between this algorithm and spatial approach in this research proposes a novel workflow of practical infrastructure monitoring in mining industry.