Multi-scale fusion network for coal mine drill rod counting based on directional object detection in complex scenes

With the advancement of intelligent mining, automated drill rod counting based on computer vision has become a crucial means to improve mining efficiency and safety. However, challenges such as dim lighting, small target sizes, diverse object perspectives, and complex visual interference in coal min...

Full description

Saved in:
Bibliographic Details
Main Authors: Fukai Zhang, Shuo Zhao, Haiyan Zhang, Yongqiang Ma, Qiang Zhang, Shaopu Wang, Wenjing Chang
Format: Article
Language:English
Published: Elsevier 2025-09-01
Series:Results in Engineering
Subjects:
Online Access:http://www.sciencedirect.com/science/article/pii/S2590123025029391
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:With the advancement of intelligent mining, automated drill rod counting based on computer vision has become a crucial means to improve mining efficiency and safety. However, challenges such as dim lighting, small target sizes, diverse object perspectives, and complex visual interference in coal mine environments significantly limit the accuracy and real-time performance of existing object detection methods. To address these issues, this paper proposes Drill-oriented Network (DrillNet), a multi-scale fusion network for counting drill rod under complex coal mine conditions using oriented object detection. The model begins by creating and annotating the LDDATA dataset. DrillNet comprises two main components: the YOLO with Multi-Scale Global Context Aggregation Network (YOLO-GC) and the Drill-Count module.The core architecture of YOLO-GC integrates the WaveletPool module, C2f-EMSCP feature extraction unit, GCFPN global context fusion pyramid network, and the oriented bounding box detection head (OBBHead), thereby effectively tackling the issues of insufficient detection accuracy and robustness in challenging coal mine scenarios. Training and testing on the custom LDDATA dataset showed that the YOLO-GC-n model achieved precision and mean average precision (mAP) of 86.9 % and 79.4 %, respectively, surpassing YOLOv10n by 12.6 % in mAP and YOLOv11n-obb by 2.8 %. Finally, the Drill-Count method in DrillNet was tested across multiple views, including drilling and withdrawal phases, with experimental results showing an average drill rod counting accuracy of 98.5 %, while achieving a processing speed of 76 frames per second (FPS), meeting real-time performance requirements.
ISSN:2590-1230