A research team has developed a lightweight, structure-aware computer vision detector called LCGA-YOLO11 that accurately identifies coal and gangue under harsh underground mining conditions. Built upon the YOLO11 object detection framework, the proposed model is engineered to provide reliable online perception during automated coal handling on conveyor belt systems. In validation experiments, the detector achieved a mean average precision of 95.94 percent at an intersection-over-union threshold of 0.5.
The detector was designed to address degraded imaging, dense stacking, adhesion, and occlusion in underground coal handling.
Crucially, the engineering design preserved the underlying end-to-end detection paradigm and the established multi-scale output mechanism of the original network. Rather than altering the complete detection pipeline, the authors introduced PP-LCNet to carry out lightweight feature extraction.
The study introduced CoalGangueAM to enhance structural and boundary representation. Structure-aware features improve boundary detection under stacking, adhesion, and occlusion. Furthermore, the researchers adopted imWIoU to improve localization stability.
The dataset was constructed from Sima Coal Mine of Lu’an Group. After augmentation, the dataset comprised 7,500 coal-gangue images. The research team partitioned this collection into 6,000 images designated for model training and 1,500 images reserved exclusively for objective validation testing.
Performance evaluations confirmed that LCGA-YOLO11 achieved substantial improvements in detection accuracy while lowering computational overhead compared to the baseline network. Alongside the 95.94 percent mAP at an intersection-over-union threshold of 0.5, the model recorded a precision of 93.46 percent and a recall rate of 90.36 percent. Concurrently, the total parameter count decreased from 2.59 million down to 2.33 million. Computational demands dropped correspondingly, with floating-point operations falling from 6.44 to 5.28 giga floating-point operations, confirming significant gains in structural efficiency.
The experimental outcomes indicate that the combination of lightweight feature extraction, boundary-focused attention, and stabilized localization provides a viable technical approach for underground coal handling. By achieving higher accuracy alongside reduced parameter counts and lower computational demands, LCGA-YOLO11 supports real-time monitoring in harsh belt transport environments. The authors note that the architecture delivers an effective balance between detection accuracy and computational efficiency, offering an operational method to maintain reliable online perception despite degraded imaging, stacking, and visual occlusion in underground mines.
Subject of Research: Lightweight computer vision detection of coal and gangue in underground mining environments
Article Title: Lightweight structure-aware detection of coal and gangue under complex operating conditions
Article References: Yue, X., Zhao, Z., Zhang, P., Chen, X., Hui, Z., & Shang, W. (2026). Lightweight structure-aware detection of coal and gangue under complex operating conditions. Geomechanics and Geophysics for Geo-Energy and Geo-Resources. https://doi.org/10.1007/s40948-026-01222-7
Image Credits: AI Generated
DOI: 10.1007/s40948-026-01222-7
Keywords: coal mining, gangue detection, YOLO11, computer vision, object detection, deep learning, mine automation, real-time monitoring, Lightweight, structure-aware, detection, coal
News Source: Blake Davidson. (November 10, 2026). Lightweight Vision Model Improves Underground Coal and Gangue Detection Accuracy. Scienmag.



