卷积神经网络在深部矿产资源预测中的应用

The Application of Convolutional Neural Network in Deep Mineral Resource Prediction

摘要: 近年来，深度学习技术在矿产资源预测领域逐渐展现出强大的潜力，研究者们不断提出并应用新的算法模型以提高预测的准确性。然而，传统的矿产资源预测模型如神经网络存在计算复杂度高、预测时间长以及容易受到噪声干扰等缺点。因此，利用深度学习方法来提高矿产资源预测的效率和精度，已成为近年来研究的热点问题之一。尽管基于深度学习的矿产资源预测研究取得了一定的进展，但仍存在一些问题，例如难以实现全过程自动化、一些深度学习方法不能有效提取矿山的空间特征以及不同类型矿产资源预测工作具有不同的数据分布特征。针对这些问题，本文提出了一个基于卷积神经网络(CNN)的深部矿产预测模型。CNN模型主要由卷积神经网络、损失函数和训练算法三个部分组成。本文以早子沟金矿的勘测数据为例，通过采用损失函数对所有已知矿体进行训练，再利用训练后的卷积神经网络对未知矿体进行预测，并利用Voxler软件将输出层结果进行立体可视化，圈定了两个潜在的靶区。实验结果显示，该模型的准确率达到了95%，表明CNN模型能够有效地提取矿体的空间特征，并在矿山空间下进行矿产资源预测，为深度矿产资源预测提供了一种新的方法，同时也为复杂矿山的矿产资源评价提供了一种新的思路。

Abstract: In recent years, deep learning technology has gradually shown strong potential in the field of mineral resource prediction, and researchers have continuously proposed and applied new algorithm models to improve the accuracy of predictions. However, traditional mineral resource prediction models such as neural networks have disadvantages such as high computational complexity, long prediction time, and susceptibility to noise interference. Therefore, using deep learning methods to improve the efficiency and accuracy of mineral resource prediction has become one of the hot research topics in recent years. Although research on mineral resource prediction based on deep learning has made some progress, there are still some problems, such as difficulty in achieving full process automation, some deep learning methods cannot effectively extract spatial features of mines, and different types of mineral resource prediction work have different data distribution characteristics. This article proposes a deep mineral prediction model based on Convolutional Neural Network (CNN) to address these issues. The CNN model mainly consists of three parts: convolutional neural network, loss function, and training algorithm. This article takes the survey data of Zaozigou Gold Mine as an example, trains all known ore bodies using a loss function, predicts unknown ore bodies using the trained convolutional neural network, and uses Voxler software to visualize the output layer results in three dimensions, delineating two potential target areas. The experimental results showed that the accuracy of the model reached 95%, indicating that the CNN model can effectively extract the spatial features of the ore body and predict mineral resources in the mining space, providing a new method for deep mineral resource prediction and a new approach for mineral resource evaluation in complex mines.