Scholarlink

<p>Accurate identification of minerals in thin-section petrographic images is fundamental in geology and petrology studies. Traditional optical microscopy techniques, while effective, are time-consuming and require expert interpretation, which can introduce subjectivity and limit efficiency. This study presents an Innovative approach employing the YOLO (You Only Look Once) deep learning algorithm for rapid and precise AI-based detection of plagioclase minerals in microscopic thin section images under cross-polarized light (XPL). A dataset comprising 92 high-resolution images was augmented to 1097 samples using data augmentation techniques to enhance model robustness. Transfer learning was utilized to leverage pre-trained YOLO weights, enabling efficient training despite limited data. The model was rigorously evaluated using standard metrics, including Precision, Recall, F1-score, and mean Average Precision (mAP), achieving high accuracy with Precision and Recall exceeding 0.98 and mAP-0.5 above 0.99 for both bounding box and mask predictions. Qualitative analysis from a geological perspective confirmed the model’s ability to reliably identify plagioclase regions across a range of petrographic textures, such as granular, intergrowth, and twinned structures, as well as under varying illumination conditions in cross-polarized light. The proposed method serves as an efficient solution for automated mineral identification, substantially accelerating petrographic analysis and minimizing human bias. This study establishes a foundation for future advancements in automated mineralogy, with promising extensions toward multi-mineral classification and the integration of multispectral imaging techniques.</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

Using the You only Look once Framework for Enhanced Plagioclase Classification in Petrographic Thin Sections

  • 期刊论文

  • 发表日期：

    2026年2月26日

查看全文
  • Ali Nazarkhanloo,
  • Andisheh Alimoradi,
  • Saeedeh Senemari,
  • Hassanreza Ghasemitabar,
  • Sajjad Talesh Hosseini

摘要

Accurate identification of minerals in thin-section petrographic images is fundamental in geology and petrology studies. Traditional optical microscopy techniques, while effective, are time-consuming and require expert interpretation, which can introduce subjectivity and limit efficiency. This study presents an Innovative approach employing the YOLO (You Only Look Once) deep learning algorithm for rapid and precise AI-based detection of plagioclase minerals in microscopic thin section images under cross-polarized light (XPL). A dataset comprising 92 high-resolution images was augmented to 1097 samples using data augmentation techniques to enhance model robustness. Transfer learning was utilized to leverage pre-trained YOLO weights, enabling efficient training despite limited data. The model was rigorously evaluated using standard metrics, including Precision, Recall, F1-score, and mean Average Precision (mAP), achieving high accuracy with Precision and Recall exceeding 0.98 and mAP-0.5 above 0.99 for both bounding box and mask predictions. Qualitative analysis from a geological perspective confirmed the model’s ability to reliably identify plagioclase regions across a range of petrographic textures, such as granular, intergrowth, and twinned structures, as well as under varying illumination conditions in cross-polarized light. The proposed method serves as an efficient solution for automated mineral identification, substantially accelerating petrographic analysis and minimizing human bias. This study establishes a foundation for future advancements in automated mineralogy, with promising extensions toward multi-mineral classification and the integration of multispectral imaging techniques.