Research on drill rod counting method in mines based on drill withdrawal process detection

In view of the problems of high false detection rate and low efficiency of existing drill rod counting methods in mines, a feature encoding counting method is designed, which mainly counts the number of drill rods actually driven in by processing and analyzing the results of the drill rig withdrawal process detection, which can effectively improve the counting accuracy and work efficiency. Due to the complex environment in the mine, the images obtained by video surveillance are easily affected by factors such as noise and lighting, resulting in the existing target detection algorithms having problems such as difficulty in feature extraction and low recognition rate. In addition, these algorithm models are highly complex and computationally intensive, which is not conducive to deployment on the edge. To address these problems, an improved YOLOv8n drill rig withdrawal process detection algorithm, called YOLOv8n–SDM, is proposed. Firstly, a new spatial pyramid pooling module is designed to enhance the feature extraction ability of the model and reduce the interference of the complex background in the mine on the drill withdrawal detection. Then, a feature aggregation module with two transformation structures is proposed to replace the C2f module in the original model, which can effectively reduce the complexity and calculation of the model while obtaining rich feature information. Finally, in order to further improve the model's ability to recognize irregular features, a multi-scale fusion detector is designed to improve the model's recognition rate of targets such as drilling rigs. Experimental results show that compared with the original YOLOv8n algorithm, the precision, recall rate, [email protected] and [email protected]–0.95 values of the improved YOLOv8n–SDM algorithm are increased by 2.7%, 2.6%, 2.2% and 1% respectively, and the model parameters, storage space occupied and GFLOPs values are reduced by 32.2%, 30.2% and 31.7% respectively. Compared with the mainstream algorithm, it can achieve more accurate detection of the drill withdrawal process in complex mine scenes, meeting the needs of actual deployment and application.