基于 ResNet-18 的三维成矿预测方法研究

目前深部隐伏矿床成为中国东部地区主要找矿目标,利用基于卷积神经网络(convolutional neural network, CNN)的三维成矿预测方法能够更好地圈定找矿靶区,指导进一步勘探。文章以安徽省宣城市茶亭地区为研究实例,开展基于ResNet-18残差网络(residual network, ResNet)的三维成矿预测方法研究。结果表明:基于ResNet-18的深层预测模型的训练准确率为99.62%;相较于逻辑回归模型和基于LeNet-5的预测模型,基于ResNet-18的三维预测模型能够在更小的成矿远景区范围内预测出更多的矿化单元,具备更优异的预测能力,可为三维成矿预测研究提供更强大的数据综合工具。

Currently, deep-seated concealed ore deposits have become the primary exploration target in the eastern region of China. The convolutional neural network(CNN) based three-dimensional mineral prospectivity modeling methods can better delineate favorable exploration areas for mineralization and guide further exploration. This paper focuses on the Chating district in Xuancheng City, Anhui Province, and conducts research on three-dimensional mineral prospectivity modeling using ResNet-18. The results show that the deep prediction model based on ResNet-18 achieves a training accuracy of 99.62%. Compared to the logistic regression model and the prediction model based on LeNet-5, it can predict more mineralized units within a smaller mineral prospectivity area, demonstrating superior predictability. Therefore, it can provide a more powerful data integration tool for research on three-dimensional mineral prospectivity modeling.