Visual image-based coal-gangue release status recognition with Detail-WaveNet and lightweight YOLOv8n model

Abstract Accurate recognition of coal-gangue is an important prerequisite for ensuring intelligent control and stable operation in top coal caving face. The main challenges in underground application are serious harsh conditions including low light and high dust, which significantly degrade the image quality and complicate the recognition process. Current coal-gangue recognition methods often struggle with noise and lack robustness under these harsh conditions. To this end, this paper proposes a novel visual image-based coal-gangue release status recognition method with Detail-WaveNet and a lightweight YOLOv8n model coal-gangue object detection-YOLO (CG-YOLO) in top coal caving face. First, the image enhancement model Detail-WaveNet is developed, which improves the coal-gangue image clarity by introducing detail enhancement convolutions and a self-modulation feature aggregation module into WaveNet model. Then, a coal-gangue recognition model CG-YOLO is proposed based on the YOLOv8n architecture, which incorporates a high-performance GPU structure (HGNetv2) and a large separable kernel attention mechanism to significantly improve the recognition accuracy of coal-gangue in harsh environments. Finally, some simulations and experiments are carried out, and the results indicate that the proposed method significantly outperform the advanced models, enabling accurate coal-gangue recognition under harsh conditions.