Scholarlink

<p>In blasting scenarios, image backgrounds are often complex and variable, and the irregular shapes of rock fragments pose significant challenges for segmentation and size recognition. To enhance the efficiency and accuracy of rock fragment size recognition in blasting operations, optimize blasting performance, and improve construction productivity, this study proposes an enhanced YOLOv11 deep learning model, named YOLOv11-EGM. YOLOv11 represents the latest iteration in the YOLO series. This study integrates ECA, GSConv, and MSCA modules into the YOLOv11 architecture to enhance detection performance. Experimental results demonstrate that the YOLOv11m variant outperforms the n, s, m, and l configurations of YOLOv8 and YOLOv11 on the dataset employed in this study. The proposed YOLOv11-EGM achieves mAP50-95 values of 0.696 and 0.554 for the Box and Mask tasks, respectively, improving by 11.7% and 2.4% compared to the original YOLOv11 model. The FLOPs (G) and FPS values are 122.2 and 11, respectively. The YOLOv11-EGM model improves recognition accuracy while maintaining computational efficiency. It features a lightweight design and strong robustness, making it well suited for real-time recognition tasks in complex environments. Compared to other rock fragment size recognition tools, the proposed method exhibits superior robustness on the same test images.</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

Improved YOLOv11-EGM deep learning model for rock fragment identification

  • 期刊论文

  • 发表日期：

    2025年7月15日

查看全文
  • Haitao Meng,
  • Ming Tao,
  • Rendong Huang,
  • Yuanquan Xu,
  • Muhammad Burhan Memon

摘要

In blasting scenarios, image backgrounds are often complex and variable, and the irregular shapes of rock fragments pose significant challenges for segmentation and size recognition. To enhance the efficiency and accuracy of rock fragment size recognition in blasting operations, optimize blasting performance, and improve construction productivity, this study proposes an enhanced YOLOv11 deep learning model, named YOLOv11-EGM. YOLOv11 represents the latest iteration in the YOLO series. This study integrates ECA, GSConv, and MSCA modules into the YOLOv11 architecture to enhance detection performance. Experimental results demonstrate that the YOLOv11m variant outperforms the n, s, m, and l configurations of YOLOv8 and YOLOv11 on the dataset employed in this study. The proposed YOLOv11-EGM achieves mAP50-95 values of 0.696 and 0.554 for the Box and Mask tasks, respectively, improving by 11.7% and 2.4% compared to the original YOLOv11 model. The FLOPs (G) and FPS values are 122.2 and 11, respectively. The YOLOv11-EGM model improves recognition accuracy while maintaining computational efficiency. It features a lightweight design and strong robustness, making it well suited for real-time recognition tasks in complex environments. Compared to other rock fragment size recognition tools, the proposed method exhibits superior robustness on the same test images.