Convolutional neural networks with SegNet architecture applied to three-dimensional tomography of subsurface electrical resistivity: CNN-3D-ERT

SUMMARY In general, the inverse problem of electrical resistivity tomography (ERT) is treated using a deterministic algorithm to find model subsurface that can numerically match apparent data acquired at ground surface and has smooth distribution been introduced as prior information. this paper, we propose new deep learning for processing 3-D reconstruction ERT. This approach relies on approximation operator considered nonlinear function linking section input underground output. performed with large amount known obtain an accurate generalization by identifying during process set parameters assigned neural networks. To train network, models are theoretically generated geostatistical anisotropic Gaussian generator, their corresponding solving Poisson's equation. These formed in way have same size trained convolutional networks SegNet architecture containing three-level encoder decoder network ending regression layer. The encoders including convolutional, max-pooling activation operations sequentially extract main features lower resolution maps. On other side, decoders dedicated upsampling concatenating feature maps transferred from compensate loss resolution. tool successfully validated different synthetic cases particular attention how quality terms noise affects effectiveness approach.