Scholarlink

<p>The quantification of magnetic particle concentration and morphology constitutes a critical component of environmental magnetism research. Traditional methodologies predominantly rely on manual identification via microscopic imagery, a labor-intensive process susceptible to human error and unsuitable for large-scale datasets. This study targets magnetic spherules within sedimentary contexts and introduces a novel deep learning framework that integrates the YOLOv8 and U-Net models, aiming to efficiently derive quantitative metrics for the abundance and morphological characteristics of magnetic spherules. Specifically, the enhanced YOLOv8 model facilitates object detection, enabling the localization and enumeration of magnetic spherules, while the U-Net model employs image segmentation techniques to precisely extract morphological parameters, including area, particle size, and roundness. Built upon convolutional neural networks, the proposed framework demonstrates robust performance in detection tasks using SEM imagery. It establishes a high-efficiency and standardized tool for the morphological analysis of large sediment sample sets, offering valuable insights for addressing scientific inquiries in Earth and environmental sciences. Furthermore, this approach contributes to advancing research on paleoclimate reconstruction and the anthropogenic impacts on environmental systems.</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

Application of deep learning in magnetic spherule detection: a combined method of YOLOv8 and U-Net models

  • 期刊论文

  • 发表日期：

    2025年5月14日

查看全文
  • Zehao Cui,
  • Yonghong Wang

摘要

The quantification of magnetic particle concentration and morphology constitutes a critical component of environmental magnetism research. Traditional methodologies predominantly rely on manual identification via microscopic imagery, a labor-intensive process susceptible to human error and unsuitable for large-scale datasets. This study targets magnetic spherules within sedimentary contexts and introduces a novel deep learning framework that integrates the YOLOv8 and U-Net models, aiming to efficiently derive quantitative metrics for the abundance and morphological characteristics of magnetic spherules. Specifically, the enhanced YOLOv8 model facilitates object detection, enabling the localization and enumeration of magnetic spherules, while the U-Net model employs image segmentation techniques to precisely extract morphological parameters, including area, particle size, and roundness. Built upon convolutional neural networks, the proposed framework demonstrates robust performance in detection tasks using SEM imagery. It establishes a high-efficiency and standardized tool for the morphological analysis of large sediment sample sets, offering valuable insights for addressing scientific inquiries in Earth and environmental sciences. Furthermore, this approach contributes to advancing research on paleoclimate reconstruction and the anthropogenic impacts on environmental systems.