基于形状上下文的成矿构造形态控矿特征深度学习及三维成矿预测——以大尹格庄金矿床为例

Study on Deep Learning of Ore-Controlling Characteristics of Geological Morphology and 3D Metallogenic Prediction Based on Shape Context—A Case Study of Dayingezhuang Gold Deposit

随着地质找矿工作逐步向深部空间的发展,深部矿、隐伏矿已成为找矿的主要对象。在对深部矿、隐伏矿的预测中,有效的成矿信息提取能够保证预测结果的准确性和可靠性。本文提出一种基于形状上下文的成矿构造形态控矿特征深度学习方法,以大尹格庄金矿床为研究对象,利用胶西北招平断裂带大尹格庄断离面趋势-起伏因素(waF、wbF)、断离面坡度因素(gF)、断离面陡缓转换部位综合场因素(fV)封装的三维形状上下文特征,利用深度学习中的图卷积神经网络,学习获得深层次的地质体形态控矿特征。研究表明,图卷积网络可提取更完备的地质体形态控矿特征,建立具有较好预测准确性的三维成矿预测模型。

With the gradual development of geological prospecting to deep space, deep ore and concealed ore have become the main objects of ore prospecting. In the prediction of deep and concealed deposits, the effective extraction of metallogenic information can ensure the accuracy and reliability of the prediction results. This paper proposes a deep learning method based on the shape context of geological morphology based on shape context. Using the 3D shape context features encapsulated by the trend-fluctuation factors (waF, wbF), slope factor (gF), and comprehensive field factors (fV) of the Dayingezhuang gold deposit, the graph convolution neural network in deep learning is used. Learn more complete morphological ore-controlling characteristics of geological bodies. Research shows that the graph convolutional network can extract more complete geological shape ore-controlling characteristics, and establish a three-dimensional ore-forming prediction model with certain prediction accuracy.