Scholarlink

<p>Foreign body detection on conveyor belts in coal mines is essential for achieving intelligent mining operations. To address the issues of high model complexity and excessive computational demands in existing object detection algorithms, a lightweight real-time detection model, GLCM-YOLO, based on an improved YOLOv11n architecture, is proposed. While maintaining detection accuracy, the parameter count and computational overhead are significantly reduced, ensuring adaptability for edge device deployment.A novel GRES_Backbone network integrating GSConv and RES Block modules is designed. The RES Block combines depthwise separable convolutions with an Efficient Channel Attention (ECA) mechanism, thereby reducing model complexity and enhancing feature extraction. To streamline the architecture further, the small object detection head is removed, reducing the neck network load and enhancing inference efficiency. To mitigate errors in small object detection, a Collaborative Augmented Detection Head (CADH) is introduced, which improves feature fusion and output through shared convolutions, thereby increasing accuracy and robustness. Additionally, an optimized Focaler-MPDIoU loss function is proposed to enhance localization of slender targets and address category imbalance.Evaluation on the test dataset reveals that the GLCM-YOLO model reaches a mean Average Precision (mAP) of 94.4%, while keeping the overall model footprint to a compact 3.6&#xa0;MB.Compared to the original YOLOv11n, the mAP improves by 4.8% points, while model parameters, FLOPs, and volume are reduced by 31.39%, 50.79%, and 32.07%, respectively. GLCM-YOLO also demonstrates strong generalization across various real-world datasets, providing robust support for intelligent and safe underground coal mine operations.</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

GLCM-YOLO: A lightweight real-time detection model for foreign body identification in coal mines

  • 期刊论文

  • 发表日期：

    2025年9月10日

查看全文
  • Zhenguan Cao,
  • Minghui Wang,
  • Lei Wang,
  • Yongjie Wang,
  • Mingming Tao

摘要

Foreign body detection on conveyor belts in coal mines is essential for achieving intelligent mining operations. To address the issues of high model complexity and excessive computational demands in existing object detection algorithms, a lightweight real-time detection model, GLCM-YOLO, based on an improved YOLOv11n architecture, is proposed. While maintaining detection accuracy, the parameter count and computational overhead are significantly reduced, ensuring adaptability for edge device deployment.A novel GRES_Backbone network integrating GSConv and RES Block modules is designed. The RES Block combines depthwise separable convolutions with an Efficient Channel Attention (ECA) mechanism, thereby reducing model complexity and enhancing feature extraction. To streamline the architecture further, the small object detection head is removed, reducing the neck network load and enhancing inference efficiency. To mitigate errors in small object detection, a Collaborative Augmented Detection Head (CADH) is introduced, which improves feature fusion and output through shared convolutions, thereby increasing accuracy and robustness. Additionally, an optimized Focaler-MPDIoU loss function is proposed to enhance localization of slender targets and address category imbalance.Evaluation on the test dataset reveals that the GLCM-YOLO model reaches a mean Average Precision (mAP) of 94.4%, while keeping the overall model footprint to a compact 3.6 MB.Compared to the original YOLOv11n, the mAP improves by 4.8% points, while model parameters, FLOPs, and volume are reduced by 31.39%, 50.79%, and 32.07%, respectively. GLCM-YOLO also demonstrates strong generalization across various real-world datasets, providing robust support for intelligent and safe underground coal mine operations.