Abstract: The coal seam of fully mechanized mining face has a complex trend.The "one-size-fits-all" mining method will accelerate the wear of roller pick but reduce the mining efficiency.Accurate identification of coal and rock is the key to solve such problems and realize intelligent mining.In this paper,the deep learning target detection algorithm YOLOv2 based on regression equation combined with the linear imaging model was used to identify and locate the coal and rock images collected underground.Meanwhile,the recognition results of coal and rock images were compared with those of Faster R-CNN and SSD.The results showed that the recognition accuracy of YOLOv2 for coal and rock is around 78%,the detection speed reaches 63 frame/s,its accuracy is 7.7% and 4.7% higher than Faster R-CNN and SSD.Also the detection speed is 763% and 40% faster than those two algorithms.The relative error between the calculated coordinates of the boundary corner points of coal seam calibrated by YOLOv2 in the Mine Survey Coordinate system and the measured coordinates is between 3.0% and 4.5%.Since the relative error is small,the mining efficiency will not be affected.The results show that YOLOv2 can recognize coal and rock accurately and quickly.