Small Target Detection Model for Open-Pit Coal Mines Based on UAV Perspective

Open-pit coal mining environments pose significant challenges for accurate detection of small and dynamic objects due to complex backgrounds, varying scales, and frequent occlusions. This study proposes an advanced object detection framework tailored for UAV perspectives to enhance both accuracy and real-time monitoring capabilities in open-pit mines. The approach integrates YOLOV11-x Enhanced, RT-DETR with ResNet101 backbone, and a hybrid Weighted Box Fusion (WBF) ensemble to effectively detect and classify critical mining elements, including company equipment, dump trucks, soil dumping activities, dust, and truck loading states. The system addresses challenges such as small target visibility, diverse object orientations, and rapid scene changes, providing robust performance in operational mining scenarios. Experimental evaluations demonstrate that the proposed multi-model strategy significantly improves detection precision, recall, and overall reliability compared to individual detection models. The framework supports real-time decision-making, enabling safer and more efficient mining operations through automated UAV monitoring.

Keywords: UAV, open-pit coal mine, small object detection, YOLOV11-x Enhanced, RT-DETR ResNet101, hybrid Weighted Box Fusion, real-time monitoring, dump truck detection, mining safety, multi-class object detection.