Scholarlink

<p>Mineral prospectivity mapping (MPM) is critical for targeting mineral exploration and mitigating associated risks. In recent studies, technological advances have been combined with deep learning techniques, such as convolutional neural network (CNN) modeling, to effectively identify and extract key correlations between 3D predictive maps and mineralization processes. Nevertheless, existing researches in this field often neglect the multi-scale characteristics of geological structures and fail to tackle the considerable computational costs tied to CNN models, thus limiting their practical use in mineral prospectivity mapping. To overcome these limitations, this paper introduces an Inception-enhanced 3D CNN model integrated with a lightweight attention mechanism (LAM) for MPM. Through the incorporation of Inception and LAM, this network’s ability to capture multi-scale geological characteristics and outline critical predicted areas is notably improved compared to the basic CNN approach. To verify the efficacy of this proposed model, 3D MPM was conducted in the Pulang porphyry Cu deposit, in southwest China. The results indicated that this proposed CNN model displays remarkable robustness and favorable generalization capabilities, successfully outlining prospective targets in the deep and the edge of the mining area and offering valuable guidance for follow-up exploration efforts. Moreover, the evaluation indices, ROC curves, and ore-controlling rate curves obtained from the modeling process show that the Inception-enhanced 3D CNN with LAM surpasses alternative models, such as the CNN model with convolutional block attention module (CBAM), CNN model with Inception module, basic CNN, support vector machine (SVM), and random forest (RF) algorithms. The results demonstrate that the proposed CNN model is capable of capturing&#xa0;3D spatial characteristics from the prospectivity prediction maps efficiently and improving its efficiency in 3D MPM better than the other methods, indicating that this method provides a favorable means to accurately identify exploration targets associated with deeply buried mineralization in future exploration surveys.</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

Multi-Scale 3D Convolution Neural Network with Lightweight Attention Mechanisms for Mineral Prospectivity Mapping in Pulang Porphyry Deposit, Yunnan Province, Southwest China

  • 期刊论文

  • 发表日期：

    2026年7月10日

查看全文
  • Xiaochen Wang,
  • Qinglin Xia,
  • Shuai Leng,
  • Yuqi Liang

摘要

Mineral prospectivity mapping (MPM) is critical for targeting mineral exploration and mitigating associated risks. In recent studies, technological advances have been combined with deep learning techniques, such as convolutional neural network (CNN) modeling, to effectively identify and extract key correlations between 3D predictive maps and mineralization processes. Nevertheless, existing researches in this field often neglect the multi-scale characteristics of geological structures and fail to tackle the considerable computational costs tied to CNN models, thus limiting their practical use in mineral prospectivity mapping. To overcome these limitations, this paper introduces an Inception-enhanced 3D CNN model integrated with a lightweight attention mechanism (LAM) for MPM. Through the incorporation of Inception and LAM, this network’s ability to capture multi-scale geological characteristics and outline critical predicted areas is notably improved compared to the basic CNN approach. To verify the efficacy of this proposed model, 3D MPM was conducted in the Pulang porphyry Cu deposit, in southwest China. The results indicated that this proposed CNN model displays remarkable robustness and favorable generalization capabilities, successfully outlining prospective targets in the deep and the edge of the mining area and offering valuable guidance for follow-up exploration efforts. Moreover, the evaluation indices, ROC curves, and ore-controlling rate curves obtained from the modeling process show that the Inception-enhanced 3D CNN with LAM surpasses alternative models, such as the CNN model with convolutional block attention module (CBAM), CNN model with Inception module, basic CNN, support vector machine (SVM), and random forest (RF) algorithms. The results demonstrate that the proposed CNN model is capable of capturing 3D spatial characteristics from the prospectivity prediction maps efficiently and improving its efficiency in 3D MPM better than the other methods, indicating that this method provides a favorable means to accurately identify exploration targets associated with deeply buried mineralization in future exploration surveys.