Extraction of Open-PIT Mine Reclamation Area with Convolutional Neural Network

Reclamation of land-damaged areas in open-pit mines is an important means of ecological restoration in mining areas. In order to effectively monitor the reclamation of mining areas, we propose an improved Mask R-CNN method to extract reclamation areas. Since traditional artificial methods and semantic segmentation networks segment the whole image, it leads to more false detections. Mask R-CNN based on instance segmentation avoids the segmentation of the whole image. Moreover, we improve its mask branch to raise accuracy. An open-pit mining area in Ordos City is chosen to be our research area and we use the GF-1 image to make open-pit mine reclamation areas dataset. Similarly, classic semantic segmentation networks are used for comparison, and verified our improved algorithm on the data of GF-1 and JL-1. In comparing with the sub-optimal algorithms, F1 scores have been improved by 1.49% and 2.3% respectively.