基于深度学习U—Net网络的地质目标识别——以铁矿为例

Geological Target Identification from Distribution of GeophysicalProperties Based on Deep Learning U-Net Network:A Synthetic Case Study of Identifying Iron Ores

地质大数据时代到来,大量地质数据为更准确的地质目标识别提供了可能,也为地质数据处理带来了巨大压力。深度学习技术可以对大批量地质数据进行特征学习,用概率预测结果表示地质目标的分布,为地质目标识别提供了新的思路和方法。本文采用深度学习方法,以铁矿识别为例,探索利用U-Net网络学习区域物性特征、构造条件等与地质目标的空间耦合关系,从而实现多源物性数据中铁矿地质目标的识别。经过多轮基于U-Net网络的地质目标识别测试,发现U-Net模型在铁矿识别任务中表现良好,能够从多源物性数据提取地质目标信息,实现高效的地质目标识别;U-Net模型从地质数据中识别地质目标的效果依赖于地质数据本身对地质目标与地质背景的区分能力。

In the era of geological big data,a large amount of geological data provides the possibility for more accurate geological target identification,but also brings great pressure to geological data processing.Deep learning technology can be used for feature learning of large quantities of geological data,and express the distribution of geological targets with probability prediction results,which provides new ideas and methods for geological target identification.In this paper,the deep learning method is used to explore the spatial coupling relationship between regional physical characteristics,structural conditions and geological targets by using U-Net network,so as to realize the identification of iron ore geological targets in multi-source physical data.After several rounds of geological target recognition tests based on U-Net network,it is found that the U-Net model performs well in iron ore identification tasks,and can extract geological target information from multi-source physical property data to achieve efficient geological target identification.The effect of U-Net model to identify geological targets from geological data depends on the ability of geological data to distinguish geological targets from geological background.