Appraisal of Resistivity Inversion Models With Convolutional Variational Encoder–Decoder Network

Recovering the actual subsurface electrical resistivity properties from tomography data is challenging because inverse problem nonlinear and ill-posed. This paper proposes a Variational Encoder-Decoder (VED) based network to obtain model, which maps apparent (input) true (output). Since deep learning (DL) models are highly dependent on training sets providing meaningful geological model complex, we have first developed an algorithm construct many realistic synthetic models. Our automatically constructs pseudo-section these We further computed two different neural architectures for comparison – UNet, attention UNet with without input depth encoding data. In end, compared our results traditional inversion borewell datasets collected aquifer mapping in hard rock terrain of West Medinipur district Bengal, India. A detailed qualitative quantitative evaluation reveals that VED approach most effective other approaches considered.