Scholarlink

<p>The automatic interpretation of geological structures can significantly accelerate the fieldwork stage to explore natural resources such as oil, water, and ore and aid engineering. Despite extensive developments in this area over the years, challenges persist in achieving efficient automated analysis. This study proposes a novel methodology that integrates deep learning networks, such as Fully Convolutional Networks, for semantic segmentation, with post-processing computer vision algorithms to extract linear geological structures from Unmanned Aerial Vehicle (UAV) imagery. The proposed method achieved an accuracy of 95% when combining Densenet for segmentation with post-processing techniques. The predicted segmentation was employed as a mask for binarization and line detection to precisely extract fractures. Comparisons of strike directions between the geological interpretation and the extracted structures revealed highly similar trends and behaviors. Furthermore, these results were benchmarked against traditional computer vision techniques, such as the Canny algorithm and binarization, demonstrating superior precision and notable advantages.</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

Semantic Segmentation for Automatic Extraction of Linear Geological Structures from UAV Imagery

  • 期刊论文

  • 发表日期：

    2025年8月11日

查看全文
  • Davi Bortolotti Batista,
  • Vitor Lamy Mesiano Savastano,
  • Milena Faria Pinto,
  • Gabriel Matos Araujo,
  • Diego Barreto Haddad

摘要

The automatic interpretation of geological structures can significantly accelerate the fieldwork stage to explore natural resources such as oil, water, and ore and aid engineering. Despite extensive developments in this area over the years, challenges persist in achieving efficient automated analysis. This study proposes a novel methodology that integrates deep learning networks, such as Fully Convolutional Networks, for semantic segmentation, with post-processing computer vision algorithms to extract linear geological structures from Unmanned Aerial Vehicle (UAV) imagery. The proposed method achieved an accuracy of 95% when combining Densenet for segmentation with post-processing techniques. The predicted segmentation was employed as a mask for binarization and line detection to precisely extract fractures. Comparisons of strike directions between the geological interpretation and the extracted structures revealed highly similar trends and behaviors. Furthermore, these results were benchmarked against traditional computer vision techniques, such as the Canny algorithm and binarization, demonstrating superior precision and notable advantages.