Intelligent Counting of Coal-Mine Drill Pipes via an Oriented-Detection YOLO11-OD Framework

Abstract Accurate counting of drill pipes in underground coal mines is critical for evaluating drilling progress and ensuring operational safety. Manual counting is inefficient and prone to high error rates. Meanwhile, existing vision-based methods frequently lack robustness in challenging underground conditions, such as low illumination, high dust concentration, and frequent occlusions. We propose an improved YOLO11-based detection and counting framework, termed YOLO11-OD, featuring: (i) an EfficientNetV2 backbone with MBConv and Fused-MBConv blocks, enhanced by ECA and CBAM attention; (ii) oriented bounding boxes with Kullback–Leibler Divergence (KLD) loss for improved orientation-aware detection; (iii) anchor optimization via K-means clustering to capture elongated drill pipe geometries; and (iv) a multi-condition counting strategy integrating DeepSORT tracking with displacement, velocity, and angular features. Field experiments demonstrate that YOLO11-OD achieves a mean average precision (mAP@0.5) of 81.3% and reduces the counting error rate to 7.2%, showing a significant improvement over the baseline YOLO11 (which had a mAP@0.5 of 71.5% and an error rate of 14.3%). These results highlight YOLO11-OD as an effective and robust solution for intelligent drill pipe monitoring in coal mine environments.