摘要
智能成矿预测是利用人工智能算法挖掘多源地质找矿大数据与矿床位置耦合关系以探寻潜在成矿区域。深度学习自主提取深层次成矿相关特征,有助于发现隐含的重要找矿信息。其中,卷积神经网络保留多源控矿因素局部空间信息,对探寻找矿位置具有重要作用。本文探讨了卷积神经网络应用于智能成矿预测的任务转化与主要挑战:转化为含矿、不含矿多源控矿因素图像样本的二分类任务,以含矿概率量化成矿潜力;存在成矿相关特征提取单一而不能满足多通道变量特征信息全面提取的不足。为此引出多尺度融合特征提取算法,同时从多个卷积尺度提取成矿相关特征并融合以提高特征多样性和丰富性,重点分析了通道融合与像素融合的不同多尺度融合特征提取方式并开展了对比实验。搭建了样本建立、模型构建、模型评价、模型预测的智能成矿预测流程框架,基于该框架以陕西省凤县金矿潜力预测为应用实例,对多源控矿因素数据进行多尺度成矿相关融合特征提取,构建了成矿预测模型并圈定了高成矿潜力靶区,为智能化找矿提供技术支撑。
Intelligent mineral prospectivity prediction utilizes artificial intelligence algorithms to explore the relationship between multi-source geological prospecting data and mineral deposit locations,in order to delineate potential prospecting areas of mineral resources.Deep learning methods can independently extract relevant features related to the mineralization in depth.It is helpful to discover hidden critical information for prospecting mineral resources.Especially,the convolutional neural network is highly capable for retaining local spatial information of multi-source ore-controlling factors.It plays an important role in searching prospecting targets of mineral resources.This paper discussed the task transformation and main challenges of the convolutional neural networks applied in intelligent mineral prospectivity prediction,It can be transformed into a dichotomy classification task of ore-bearing and ore-barren samples of multi-source ore-controlling factors,for quantifying the mineralization potential by using the ore-bearing probability.However,there is a shortcoming of single extraction of features related to the mineralization,which can not satisfy for the requirement of comprehensive extraction of information of multi-channel variable features.Therefore,this paper introduced a multi-scale feature fusion extraction algorithm to have extracted features related to the mineralization from multiple convolutional scales,and then these multi-scale features were fused to improve the diversity and fertility of the obtained features related to the mineralization.Especially,the difference between the channel fused and pixel fused multi-scale feature extraction algorithms was analyzed,and the comparative experiment was carried out.Finally,the framework for intelligent prediction of mineral prospecting was established with steps of the sample establishment,model construction,model evaluation and model prediction,Based on this framework,a case study of the mineral prospectivity prediction of gold deposits in Fengxian County,Shaanxi Province has been carried out by extracting multi-scale fusion features related to the mineralization from multi-source prospecting information.A predictive model has been constructed and some targeting areas with high potential of mineral prospecting have been outlined.This study can provide technical support for intelligent prospecting of mineral resources.
作者
杨娜
YANG Na(College of Computer Science and Technology,Xi'an University of Science and Technology,Xi'an 710054,China)
出处
《矿物岩石地球化学通报》
北大核心
2025年第3期478-491,共14页
Bulletin of Mineralogy, Petrology and Geochemistry
基金
陕西省自然科学基础研究计划青年项目(2024JC-YBQN-0288)。
关键词
智能成矿预测
深度学习
多尺度特征融合
卷积神经网络
intelligent mineral prospectivity prediction
deep learning
multi-scale feature fusion extraction
convolutional neural network
