Research on downhole drilling target detection based on improved Yolov8n
Abstract The drilling in underground coal mine drill sites is essential for preventing and managing gas disasters, water hazards, and hidden geological risks. It plays a vital role in strengthening the disaster prevention and control capabilities of coal mines in China. To monitor the drilling proce...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Nature Portfolio
2025-07-01
|
| Series: | Scientific Reports |
| Subjects: | |
| Online Access: | https://doi.org/10.1038/s41598-025-11689-1 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Abstract The drilling in underground coal mine drill sites is essential for preventing and managing gas disasters, water hazards, and hidden geological risks. It plays a vital role in strengthening the disaster prevention and control capabilities of coal mines in China. To monitor the drilling process in real time and enhance the efficiency of target detection at underground coal mine drill sites, an improved algorithm based on Yolov8n has been proposed, which offers advantages compared with the traditional detection methods. This detection focuses on identifying and locating key targets in the drilling sites, including five categories: grippers, drill chucks, coal miners, mine helmets, and drill pipes. The multicore initiator module C2f_PKI is employed to replace C2f as the Backbone network to accelerate target detection and reduce model complexity. By incorporating FDPN and DASI fusion modules into the Head module section, the aim is to reduce model complexity and enhance detection accuracy. The Focaler_MDPIoU loss function was introduced to replace the CIoU loss function, and combined with a lightweight MLCA attention mechanism, a new convolutional block structure was formed, thereby enabling better feature fusion and helping the model focus more centrally on the important parts of the input image. Thereby improving the recognition accuracy and generalization ability of the model, and further enhancing the detection performance. The results indicate that compared with the original model, the improved model achieves a 14% reduction in model storage size and a 17% decrease in the number of parameters. Additionally, it enhances Precision by 1.4% and increases the mean average precision (mAP) by 0.9%. This offers a technical solution for underground target recognition in complex environments. |
|---|---|
| ISSN: | 2045-2322 |