Research on coal gangue recognition and detection model based on improved YOLOv8n

Aiming at the problems of poor recognition effect and low efficiency of the coal gangue sorting task in the coal mining industry, a recognition and detection model of coal gangue is proposed based on improved YOLOv8n. Firstly, the standard convolution in YOLOv8n is replaced by Ghost convolution, and the Ghost convolution is integrated into C2f module to construct a C2fGhost module, which reduced the number of network parameters. Secondly, a multi-scale adaptive fusion module is designed to replace the splicing module of YOLOv8n, which optimized the multi-scale feature fusion strategy and improved the overall quality of feature integration as well as recognition accuracy. Finally, a dynamic detection head is adopted to enhance the model's processing capabilities in the dimensions of scale, spatiality, and task. The experimental results show that compared to the YOLOv8n model, the improved model achieves a 3% increase in average accuracy mean P, an 8.4 FPS improvement in detection frame rate, a 24% reduction in parameter count, a 16% decrease in computational load, and a 25% reduction in model size.