Scholarlink

<p>The Rock Quality Designation (RQD) system is a crucial tool in geotechnical engineering for assessing rock mass quality. This system which involves manual measurement of solid core piece equal to or exceeding 100&#xa0;mm in length is time-consuming and labor-intensive. Recently, deep learning has enabled researchers to develop and propose automated models for estimating RQD. These models, however, estimate RQD at the core box level and hence make comparison of results with manual RQD results challenging. This research proposes a novel automatic RQD analysis model, which estimates RQD at the core-run-level, using Faster Region-based Convolutional Neural Network (R-CNN) and You Only Look Once version 11 (YOLOv11). The models were trained on 160 core images which comprise of a wide variety of core tray images. The results were highly accurate Faster R-CNN-based and YOLOv11-based solid core (and other core tray features) detection models which detected features with high precision and recall (both approximating 1 for each model). Compared to manual estimates of 200 core runs, RQD analysis based on YOLOv11 proved superior to that of Faster R-CNN; with the former yielding a Mean Average Percentage error of 3.43% and the latter having an MAPE of 3.21%.</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

Automatic Solid Core Detection Using Faster R-CNN and Yolo11 for Real-Time RQD Estimation

  • 期刊论文

  • 发表日期：

    2026年6月16日

查看全文
  • Sylvanus Sebbeh-Newton,
  • George Kwame Frimpong,
  • Kwame Atta Frimpong,
  • Jamel Seidu,
  • Ezekiel Sackitey Nanor,
  • Issaka Yakubu

摘要

The Rock Quality Designation (RQD) system is a crucial tool in geotechnical engineering for assessing rock mass quality. This system which involves manual measurement of solid core piece equal to or exceeding 100 mm in length is time-consuming and labor-intensive. Recently, deep learning has enabled researchers to develop and propose automated models for estimating RQD. These models, however, estimate RQD at the core box level and hence make comparison of results with manual RQD results challenging. This research proposes a novel automatic RQD analysis model, which estimates RQD at the core-run-level, using Faster Region-based Convolutional Neural Network (R-CNN) and You Only Look Once version 11 (YOLOv11). The models were trained on 160 core images which comprise of a wide variety of core tray images. The results were highly accurate Faster R-CNN-based and YOLOv11-based solid core (and other core tray features) detection models which detected features with high precision and recall (both approximating 1 for each model). Compared to manual estimates of 200 core runs, RQD analysis based on YOLOv11 proved superior to that of Faster R-CNN; with the former yielding a Mean Average Percentage error of 3.43% and the latter having an MAPE of 3.21%.