世界核地质科学

矿物及围岩蚀变信息是铀及多金属资源勘查中识别岩浆-热液活动、构造有利部位和成矿环境的重要线索。高分五号（GF-5）卫星搭载的高光谱成像仪（AHSI）具备可见光—近红外—短波红外连续光谱观测能力，可为铁氧化物/铁羟基、铝羟基、镁羟基、碳酸盐和硫酸盐等典型矿物信息识别提供重要数据支撑。然而，在大面积多景GF-5高光谱影像实际应用中，不同景影像成像条件、地表覆盖、影像质量和背景地物差异明显，条带噪声、坏线、阴影、水体、植被和覆盖层等因素容易影响矿物信息提取效果，制约其规模化和业务化应用。针对上述问题，本文面向GF-5 AHSI反射率数据，构建了由数据接入与预检查、质量控制、有效像元掩膜、多通道端元提取、矿物目标自动筛选、矿物族级填图、类型细化和成果产品化组成的自动化处理流程。该流程综合利用PPI、VCA和N-FINDR等端元提取方法以及ACE、CEM目标检测方法，实现矿物信息从端元组织到空间填图的流程化处理。以新疆维吾尔自治区拜城、富蕴、若羌和哈密4景GF-5 AHSI影像开展流程验证，并以柳园典型验证区开展矿物信息地质分析。结果表明，新疆维吾尔自治区拜城、富蕴、若羌和哈密4景影像数据共完成36组多通道端元提取任务，自动提取端元1 728条，筛选出66个进入填图的目标端元，形成5类矿物族填图结果，说明该流程具备较好的批量处理和矿物填图结果可追溯能力。柳园验证区提取出中铝白云母、低铝白云母、方解石、白云石、绿泥石和绿帘石等矿物信息，提取的矿物信息与花岗质岩体、岩体边界、地层接触带及断裂构造之间具有一定空间对应关系。研究表明，构建的自动化流程能够实现多景GF-5影像的规范化处理和典型矿物信息自动提取，可为区域地质遥感调查、热液型铀成矿环境与蚀变异常初步筛选，以及多源综合找矿评价提供标准化矿物信息产品。

Mineral assemblages and wall-rock alteration provide important clues for identifying magmatic–hydrothermal activity，structurally favorable zones，and metallogenic environments in uranium and polymetallic exploration. The Advanced Hyperspectral Imager （AHSI）onboard the GaoFen-5 （GF-5）satellite provides continuous spectral coverage from the visible and near-infrared to the shortwave-infrared regions，offering essential data support for identifying typical minerals such as iron oxides/hydroxides，Al-bearing hydroxyl minerals，Mg-bearing hydroxyl minerals，carbonates，and sulfates. However，in large-area applications involving multiple GF-5 hyperspectral scenes，substantial differences in imaging conditions，surface cover，image quality，and background materials can occur among scenes. Factors such as striping noise，bad detector lines，shadows，water bodies，vegetation，and surficial cover may adversely affect mineral information extraction，thereby limiting its large-scale and operational application.To address these issues，an automated processing workflow was developed for GF-5 AHSI reflectance data. The workflow comprises data ingestion and preliminary inspection，quality control，valid-pixel masking，multi-channel endmember extraction，automated mineral-target screening，mineral-family mapping，mineral-type refinement，and product generation. It integrates endmember extraction methods，including the Pixel Purity Index （PPI），Vertex Component Analysis （VCA），and N-FINDR，with target detection methods such as the Adaptive Coherence Estimator （ACE）and Constrained Energy Minimization （CEM），enabling standardized processing from endmember organization to spatial mineral mapping.The workflow was validated using four GF-5 AHSI scenes covering Baicheng，Fuyun，Ruoqiang，and Hami in the Xinjiang Uygur Autonomous Region，China，and geological analysis of the extracted mineral information was conducted in the representative Liuyuan validation area. A total of 36 multi-channel endmember extraction tasks were completed for the four scenes，yielding 1 728 automatically extracted endmember spectra. Of these，66 target endmembers were selected for mineral mapping，resulting in maps of five mineral families. These results demonstrate that the proposed workflow provides effective batch-processing capability and ensures the traceability of mineral mapping results.In the Liuyuan validation area，medium-Al muscovite，low-Al muscovite，calcite，dolomite，chlorite，and epidote were identified. The spatial distributions of these minerals exhibit correspondence with granitoid intrusions，intrusion boundaries，stratigraphic contact zones，and fault structures. The results indicate that the proposed automated workflow can support the standardized processing of multiple GF-5 scenes and the automated extraction of typical mineral information. It can provide standardized mineral information products for regional geological remote-sensing surveys，the preliminary screening of hydrothermal uranium metallogenic environments and alteration anomalies，and integrated multi-source mineral exploration assessment.