Accurate foreign object detection for the coal preparation industry based on computer vision and deep learning algorithms
Kefei Zhang Teng Wang Liang Xu Jesse Thé Zhongchao Tan Hesheng Yu a Key Laboratory of Coal Processing and Efficient Utilization of Ministry of Education,School of Chemical Engineering and Technology,China University of Mining and Technology,Xuzhou,Jiangsu,Chinab Department of Mechanical & Mechatronics Engineering,University of Waterloo,Waterloo,Ontario,Canadac Lakes Environmental Research Inc,Waterloo,Ontario,Canadad Eastern Institute for Advanced Study,Eastern Institute of Technology,Ningbo,Zhejiang,Chinae Department of Energy and Power Engineering,Tsinghua University,Beijing,China
DOI: https://doi.org/10.1080/19392699.2024.2434891
2024-12-04
International Journal of Coal Preparation and Utilization
Abstract:Accurate detection of foreign objects during industrial coal preparation is crucial to ensuring the safety of equipment and personnel, as well as maintaining the green utilization of the coal product. The complexity of industrial coal preparation environments presents challenges for vision-based foreign object detection. This work introduces the SHA-DH-YOLOX algorithm, designed to enhance detection accuracy. The proposed algorithm boosts three notable improvements. First, the Shuffle Attention mechanism (SHA) is integrated into the YOLOX backbone to strengthen the feature extraction of essential information from input images. Second, the Dynamic head (Dyhead) is introduced into the feature fusion to enhance the detector's representation capability, improving scale-, spatial-, and task-awareness. Third, the original Intersection over Union (IoU) loss function is replaced with SCYLLA-IoU (SIoU) to achieve more accurate bounding boxes and enhanced training stability. These improvements work collaboratively with YOLOX-M, resulting in it outperforming state-of-the-art baseline models, including YOLOv5, YOLOv6, YOLOv7, YOLOv8, and original YOLOX. The developed SHA-DH-YOLOX algorithm improves AP 50 by 1.87 to 7.97% compared to baseline models of equivalent size. Robustness tests further affirm the stability of the SHA-DH-YOLOX model when facing diverse and challenging scenarios. This pioneering work provides valuable tools for achieving safe and reliable coal preparation practices.
energy & fuels,mining & mineral processing