A Novel RSPD-YOLOv8 Algorithm in the Detection of Coal Mining Industry

This paper proposes a new model that combines object detections algorithms and motion state detection algorithms to address the issues related to the detection of “three violations” in coal mining industry. Based on the core of YOLOv8, the novel RSPD-YOLOv8 network was constructed by reducing the stride of the convolutional layers in the neck and backbone and introducing the Reduce-SPD module. The background subtraction algorithm and a multi-object tracking algorithm (Bot-SORT) are integrated into the model RSPD-YOLOv8. Furthermore, image enhancement techniques are used on existing training samples to increase training data, improving the model's generalization, and lowering the risk of overfitting. Experimental results show that this method achieved a mAP50 of 97.6% and a mAP50-95 of 66.9% on a self-built dataset containing five categories of data from underground coal mines. Compared to the original YOLOv8n model, this represents an improvement of 1.4 percentage points in mAP50 and 2.5 percentage points increase in mAP50-95.