标题： 使用递归卷积神经网络方法的地质建模 显示英文标题

标题： Geological modeling using a recursive convolutional neural networks approach

摘要： 资源模型受到地质单元范围的限制，而这些地质单元通常取决于岩石类型、蚀变和矿化作用。 必须从钻孔提供的稀缺信息以及关于矿床所在地质环境的有限理解中构建这些地质单元的三维模型。 在这项工作中，我们提出了一种基于递归卷积神经网络方法（RCNN）的新技术，用于多点地质统计模拟。 该方法需要条件数据和训练图像，训练图像描述了预期在矿床中发现的地质结构类型。 此训练图像是用来学习所发现的类别模式，并且这些模式会在最终的模拟模型中根据实际钻孔样本记录期间发现的类别加以约束。 在智利的一个铜矿床中进行了一个岩性建模过程，以展示该方法。 通过与现有技术的比较以及空间度量标准来阐明概念和RCNN属性。 此外，通过对理论视角的简要回顾并探讨其某些实用方面，讨论了该方法论的优势和劣势。 显示英文摘要

摘要： Resource models are constrained by the extent of geological units that often depend on the lithology, alteration and mineralization. A three dimensional model of these geological units must be built from scarce information coming from drillholes and limited understanding about the geological setting in which the ore deposit is places. In this work, we present a new technique for multiple-point geostatistical simulation based on a recursive convolutional neural network approach (RCNN). The method requires conditioning data and a training image that depicts the type of geological structures expected to be found in the deposit. This training image is used to learn the patterns of categories found and these are imposed in the final simulated model conditioned by the categories found during logging at the actual drill-hole samples. A lithological modeling process is carried out in a copper deposit in Chile to demonstrate the method. Comparison with current techniques and spatial metrics are used to clarify concepts and RCNN properties. Also, strengths and weaknesses of the methodology are discussed by briefly reviewing the theoretical perspective and looking into some of its practical aspects.