Scholarlink

This study focuses on the deep prospecting problems in the Jiapigou gold mining area in Jilin Province. In order to address the bottlenecks of traditional geophysical exploration in the Archean greenstone belt, such as the limited detection depth (<500 m), the model analysis error of multi-period structural superposition area exceeding 35%, and the insufficient extraction of mineralization information in vegetation-covered areas by satellite remote sensing. We innovatively construct a multi-source remote sensing fusion framework based on WorldView-3 hyperspectral (band 400–2500 nm), TerraSAR-X radar (resolution 1 m) and UAV LiDAR point cloud (density 28pt/m2). Through terrain correction (assisted by SRTM DEM) and band optimization, vegetation interference is eliminated (information loss rate < 12% in areas with NDVI > 0.6); a three-dimensional feature matrix containing 12 types of altered mineral spectral libraries (sericite Al–OH absorption depth 0.32 ± 0.05) and structural line density (5.8 km/km2) is established; a hybrid graph convolutional network (HGCN) is developed, integrating the Inception v3 module to extract spatial-spectral features and introducing transfer learning (pre-trained weights came from the Jiaodong gold mining area); and Monte Carlo Dropout is combined to quantify prediction uncertainty. The verification results show that in the application model of the northwest-trending ductile shear zone of Jiapigou, the F1-score reached 0.889, the precision rate is 93.5%, and three hidden ore bodies at a depth of 800–1500 m are successfully delineated (verified by CSAMT and drilling, the cumulative thickness of Au mineralization is 38.6 m, and the highest grade is 3.2 g/t). This study provides a high-precision technical solution for the second prospecting space detection in the deep ancient orogenic belt.

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

Research on Deep Mineral Prediction Model Integrating Remote Sensing Data and Machine Learning Algorithm

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Jun Gu

摘要

This study focuses on the deep prospecting problems in the Jiapigou gold mining area in Jilin Province. In order to address the bottlenecks of traditional geophysical exploration in the Archean greenstone belt, such as the limited detection depth (<500 m), the model analysis error of multi-period structural superposition area exceeding 35%, and the insufficient extraction of mineralization information in vegetation-covered areas by satellite remote sensing. We innovatively construct a multi-source remote sensing fusion framework based on WorldView-3 hyperspectral (band 400–2500 nm), TerraSAR-X radar (resolution 1 m) and UAV LiDAR point cloud (density 28pt/m2). Through terrain correction (assisted by SRTM DEM) and band optimization, vegetation interference is eliminated (information loss rate < 12% in areas with NDVI > 0.6); a three-dimensional feature matrix containing 12 types of altered mineral spectral libraries (sericite Al–OH absorption depth 0.32 ± 0.05) and structural line density (5.8 km/km2) is established; a hybrid graph convolutional network (HGCN) is developed, integrating the Inception v3 module to extract spatial-spectral features and introducing transfer learning (pre-trained weights came from the Jiaodong gold mining area); and Monte Carlo Dropout is combined to quantify prediction uncertainty. The verification results show that in the application model of the northwest-trending ductile shear zone of Jiapigou, the F1-score reached 0.889, the precision rate is 93.5%, and three hidden ore bodies at a depth of 800–1500 m are successfully delineated (verified by CSAMT and drilling, the cumulative thickness of Au mineralization is 38.6 m, and the highest grade is 3.2 g/t). This study provides a high-precision technical solution for the second prospecting space detection in the deep ancient orogenic belt.