The study of coal gangue segmentation for location and shape predicts based on multispectral and improved Mask R-CNN

The intelligent separation of gangue is of great significance to the clean and efficient utilization of coal. We propose an improved Mask R-CNN combined with multispectral imaging for coal gangue instance segmentation. Based on the lightweight of the backbone network and neck network to improve the classic mask R-CNN, recorded as L-Mask R-CNN. The positioning precision of the improved Mask R-CNN for coal and coal gangue is 96.22% and 95.12%, respectively, and the test time consumption is 6.436 s. Besides, compared to YOLO v4 and CenterNet, U-Net, and Deeplab v3+, the L-Mask R-CNN can more precisely obtain the 2D shape of each gangue instance, which allows us to evaluate its relative size. The results show that the improved L-Mask R-CNN can accurately locate the coal gangue and allows to get its relative size, which is of great significance to the intelligent separation of coal gangue. Segmentation results of coal and coal gangue in multispectral images. • Instance segmentation of coal gangue is based on multispectral imaging. • Improve the classic Mask RCNN to segment coal gangue. • Design a lightweight backbone network to improve the segmentation performance of coal gangue. • Realize the accurate positioning, shape prediction, and relative size evaluation of coal gangue.