Scholarlink

<p>Many&#xa0;applications in carbonate facies analysis require a compositional analysis of the grain, cement, and pore types.&#xa0;Image analysis based on deep learning models can help automatically extract features for identifying objects and interpreting carbonate thin sections. However, small objects are detected with a lower average precision than medium and large objects due to the loss of information during the deep convolution operation. Existing object detection algorithms cannot simultaneously achieve a high detection accuracy and detection speed, which hinders the further study of petroliferous carbonate successions. You Only Look Once version 5 (YOLOv5) is an advanced, fast, and accurate detector. In this study, the applicability and performance of YOLOv5-based object detection approaches were assessed by conducting a carbonate compositional analysis. The training data comprised more than 6800 individually labeled objects from 1000 carbonate petrographic images. The dataset was grouped into nine different classes for the object detection tasks. Even with a small amount of training, the YOLOv5 could achieve a precision of 99.0%, a recall rate of 98.4%, and a mean average precision of 91.6% for object detection by combining the scale sequence feature pyramid network. The study not only meets the accuracy requirements of identifying multi-scale objects, particularly small objects, but also meets the detection speed requirement, with a significant application potential in identifying carbonate components.</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

Intelligent identification of carbonate components based on deep learning

  • 期刊论文

  • 发表日期：

    2025年1月10日

查看全文
  • Yuqing Dong,
  • Chenlin Hu,
  • Jonathan Atuquaye Quaye,
  • Ning Lu,
  • Lingfeng Zhao

摘要

Many applications in carbonate facies analysis require a compositional analysis of the grain, cement, and pore types. Image analysis based on deep learning models can help automatically extract features for identifying objects and interpreting carbonate thin sections. However, small objects are detected with a lower average precision than medium and large objects due to the loss of information during the deep convolution operation. Existing object detection algorithms cannot simultaneously achieve a high detection accuracy and detection speed, which hinders the further study of petroliferous carbonate successions. You Only Look Once version 5 (YOLOv5) is an advanced, fast, and accurate detector. In this study, the applicability and performance of YOLOv5-based object detection approaches were assessed by conducting a carbonate compositional analysis. The training data comprised more than 6800 individually labeled objects from 1000 carbonate petrographic images. The dataset was grouped into nine different classes for the object detection tasks. Even with a small amount of training, the YOLOv5 could achieve a precision of 99.0%, a recall rate of 98.4%, and a mean average precision of 91.6% for object detection by combining the scale sequence feature pyramid network. The study not only meets the accuracy requirements of identifying multi-scale objects, particularly small objects, but also meets the detection speed requirement, with a significant application potential in identifying carbonate components.