Scholarlink

<p>In the realm of intelligent mine construction, the efficient separation of gangue from raw coal plays a pivotal role in enhancing coal resource utilization and promoting environmental sustainability. Traditional methods face challenges in real-time and high-accuracy detection amidst coal dust interference, high-speed motion, and uneven lighting conditions on conveyor belts. To address these issues, this paper introduces an innovative, lightweight detection network based on YOLOv9, incorporating a multi-path attention mechanism with large kernels to enhance feature extraction under challenging conditions. The proposed network also features a bottom-up path structure for reduced semantic information loss and an adaptive frequency decomposition module for improved representation of edge and texture details. Furthermore, a Ghost-InceptionV2 convolution module is integrated into the detection head, achieving a lightweight design without compromising accuracy. Extensive experiments demonstrate that our network significantly outperforms traditional approaches, achieving an 11% accuracy improvement over the baseline while maintaining real-time performance at 97 FPS. These findings underscore the practical applicability of our framework for real-time gangue detection in complex mining environments. Code is available at: <a href="https://github.com/chrales2048/HMA-yolo">https://github.com/chrales2048/HMA-yolo</a>.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhancing gangue recognition in coal mines: a lightweight network with multi-path attention

  • 期刊论文

  • 发表日期：

    2026年3月18日

查看全文
  • Zheng Wang,
  • Le Peng,
  • Yujiang Liu

摘要

In the realm of intelligent mine construction, the efficient separation of gangue from raw coal plays a pivotal role in enhancing coal resource utilization and promoting environmental sustainability. Traditional methods face challenges in real-time and high-accuracy detection amidst coal dust interference, high-speed motion, and uneven lighting conditions on conveyor belts. To address these issues, this paper introduces an innovative, lightweight detection network based on YOLOv9, incorporating a multi-path attention mechanism with large kernels to enhance feature extraction under challenging conditions. The proposed network also features a bottom-up path structure for reduced semantic information loss and an adaptive frequency decomposition module for improved representation of edge and texture details. Furthermore, a Ghost-InceptionV2 convolution module is integrated into the detection head, achieving a lightweight design without compromising accuracy. Extensive experiments demonstrate that our network significantly outperforms traditional approaches, achieving an 11% accuracy improvement over the baseline while maintaining real-time performance at 97 FPS. These findings underscore the practical applicability of our framework for real-time gangue detection in complex mining environments. Code is available at: https://github.com/chrales2048/HMA-yolo.