Abstract: In order to improve the efficiency and precision of underground drill pipe counting in coal mines, a coal mine underground drill pipe counting method based on the improved YOLOv8n model is proposed. The YOLOv8n-TbiD is established.The model can accurately detects and segments drill pipes in mine drilling rig working videos. The main improvements include the following points. In order to effectively capture the boundary information of drill rods and improve the precision of the model in recognizing drill rod shapes, the weighted bidirectional feature pyramid network (BiFPN) is used instead of the path aggregation network (PANet). To address the issue of drill pipe objects being easily confused with dim mine environments, Triplet Attention is added to the SPPF module of the Backbone network to enhance the model's capability to suppress background interference. In response to the small proportion of drill pipes in the image and the complexity of background information, the Dice loss function is used to replace CIoU loss function to optimize the segmentation processing of drill pipe objects in the model. The method uses the YOLOv8n-TBiD model to segment the drill pipe and its mask information. A drill pipe counting algorithm is designed based on the rule that the mask area of the drill pipe decreases during drilling and suddenly increases when a new drill pipe is installed. The working video of the drilling rig in the fully mechanized working face is selected, in order to conduct experimental verification of drill pipes counting method based on YOLOv8n-TBiD model. The experimental results show that the mean average precision of the YOLOv8n-TBiD model for detecting drill pipes reaches 94.9%. Compared with the comparative experimental models GCI-YOLOv4, ECO-HC, P-MobileNetV2, YOLOv5, and YOLOX, the accuracy increases by 4.3%, 7.5%, 2.1%, 6.3%, and 5.8%, respectively, and the detection speed increases by 17.8% compared to the original YOLOv8n model. The proposed drill pipe counting algorithm achieves precision of 99.3% on video datasets from different underground coal mine environments.