基于深度学习的等值反磁通瞬变电磁数据反演

Deep Learning Inversion of Opposing-Coils Transient Electromagnetic Data

等值反磁通瞬变电磁法(OCTEM)是一种新兴的地球物理勘探方法。它在城市和工程勘测中越来越受欢迎。本研究采用深度学习(DL)反演方法来解决等值反磁通瞬变电磁反演问题。首先我们创建了一个合成样本数据集,包括50,000个电阻率模型–OCTEM响应对组成的合成样本数据集,用于网络训练。采用流行的卷积神经网络架构U-Net和残差块(Res Net)来构建我们的深度学习反演模型OCTEM InvNet。实验反演实例表明,所提出的OCTEM InvNet能产生准确可靠的反演结果。OCTEM InvNet在重建地下电阻率结构方面表现较好,并能实现瞬时反演。

Opposing-coils transient electromagnetic (OCTEM) is a burgeoning geophysical exploration method. It is gaining popularity in urban and engineering surveys due to the advantages of high efficiency and resolution. In this study, a deep learning (DL) inversion method is proposed to solve the OCTEM inverse problem. We create a synthetic sample dataset consisting of 50,000 resistivity model-OCTEM response pairs for network training and employ the popular convolutional neural network architectures U-Net and residual block (Res Net) to construct our deep learning inversion model OCTEM InvNet. The experimental inversion examples demonstrate that the proposed OCTEM InvNet can produce accurate and reliable inversion results. OCTEM InvNet performs well in reconstructing the subsurface resistivity structure and enables instantaneous inversion.