Scholarlink

<p>The Himalayan orogen, known for its two world-class leucogranite belts, presents significant potential for the exploration of rare metals, including Ta, Sn, Be, Li, and W. Remote-sensing technology, particularly satellite-based imagery, is widely used in geological surveys to address challenges posed by harsh natural environments and inaccessible terrains. However, the vast scale of the Himalayan orogen (over 130,000&#xa0;km<sup>2</sup>) and the high computational costs associated with lithological mapping using remote-sensing data have hindered effective exploration in this region. On this account, we present a novel approach utilizing the Google Earth Engine (GEE) platform combined with deep learning algorithms to overcome these challenges. A band selection strategy based on laboratory-measured spectral features and random forest algorithm was employed to efficiently screen optimal spectral band combinations from Sentinel-2 and ASTER images. Subsequently, a U-Net fully convolutional neural network (FCNN) was applied to detect the spatial distribution of leucogranites within the Himalayan orogen. The results show that the U-Net network outperformed random forest method in successfully identifying Himalayan leucogranites, with the mapped areas aligning with previously reported occurrences, demonstrating the potential of using GEE and deep learning for large-scale lithological mapping task. This study provides a scalable and efficient method for exploring rare metal resources in the Himalayan orogen and offers a feasible approach for large-scale geological surveys in remote and difficult-to-access regions. The findings have significant implications for future exploration and geological mapping using remote-sensing technologies.</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

Large-scale Himalayan Leucogranite Mapping Based on Multi-source Remote-Sensing Data and U-Net Convolutional Network

  • 期刊论文

  • 发表日期：

    2025年7月3日

查看全文
  • Dakai Guo,
  • Ziye Wang,
  • Renguang Zuo

摘要

The Himalayan orogen, known for its two world-class leucogranite belts, presents significant potential for the exploration of rare metals, including Ta, Sn, Be, Li, and W. Remote-sensing technology, particularly satellite-based imagery, is widely used in geological surveys to address challenges posed by harsh natural environments and inaccessible terrains. However, the vast scale of the Himalayan orogen (over 130,000 km2) and the high computational costs associated with lithological mapping using remote-sensing data have hindered effective exploration in this region. On this account, we present a novel approach utilizing the Google Earth Engine (GEE) platform combined with deep learning algorithms to overcome these challenges. A band selection strategy based on laboratory-measured spectral features and random forest algorithm was employed to efficiently screen optimal spectral band combinations from Sentinel-2 and ASTER images. Subsequently, a U-Net fully convolutional neural network (FCNN) was applied to detect the spatial distribution of leucogranites within the Himalayan orogen. The results show that the U-Net network outperformed random forest method in successfully identifying Himalayan leucogranites, with the mapped areas aligning with previously reported occurrences, demonstrating the potential of using GEE and deep learning for large-scale lithological mapping task. This study provides a scalable and efficient method for exploring rare metal resources in the Himalayan orogen and offers a feasible approach for large-scale geological surveys in remote and difficult-to-access regions. The findings have significant implications for future exploration and geological mapping using remote-sensing technologies.