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

To address issues such as missed detection of small targets, low computational efficiency, and difficulties in deploying on edge devices encountered in the detection of unmanned mining trucks under the complex environment of open-pit coal mines, this paper proposes a lightweight object detection model based on improved YOLOv8.The model integrates the GhostNet lightweight convolution module, achieving a 66% reduction in parameters and a 57.5% decrease in computational load while maintaining feature extraction capability. It introduces the CBAM attention mechanism to enhance the model's ability to focus on key features in complex environments and improve target discrimination accuracy. The WIoUv3 loss function is adopted to strengthen sensitivity to small targets through a dynamic non-monotonic focusing mechanism, optimizing the bounding box localization effect. Additionally, a hierarchical perception pruning strategy with layer-adaptive weight magnitude scoring is combined, which reduces parameters by 50% while stably maintaining the mAP@50 metric at 0.852.Experimental results demonstrate that the proposed model reduces inference latency to 332.15 milliseconds on edge devices, compresses storage capacity to 4.71 megabytes, and improves computational efficiency by 25.3% compared to traditional algorithms. Moreover, its detection performance outperforms Faster R-CNN, SSD, and YOLO series algorithms, providing a real-time environmental perception solution that balances accuracy and efficiency for unmanned driving systems in open-pit coal mines.