Improving 2D resistivity model obtained from DC Resistivity Data Inversion by using Convolutional Neural Network Algorithm to Find Buried Archaeological Remains

Abstract

The Direct Current Resistivity (DCR) method is one of the well-known geophysical methods used for a wide range of areas such as mining geophysics, hydrogeophysics, and archaeogeophysics investigations. DCR data is generally collected along profile using multi-electrode and multi-channel measurement systems and interpreted using ‘two-dimensional (2D) or three-dimensional (3D) inversion algorithms. The inverse problem of DCR data is ill-posed (nonlinear, nonunique, and unstable). Therefore, the generally smoothing regularization inversion method is used for DCR data inversion. Additionally, a homogenous resistivity model is used as the initial model in regularized inversion. Hence, we generally obtain a smooth resistivity model after 2D/3D inversion. However, some structures such as buried archaeological targets, cavities, and fault structures have sharp boundaries with their neighboring medium.   In this research, we propose enhancing 2D DCR data inversion results using a convolutional neural network (CNN), aiming for sharp boundaries. We developed a U-net-based CNN algorithm, named DCR2D_Net_Archeo. This method utilizes 2D inversion results as the input, with the real resistivity model serving as the output, streamlining geophysical data interpretation for archeological applications.  We tested the DCR2D_Net_Archeo algorithm by using synthetic and real data.  We showed that the developed resistivity model enhancement algorithm, DCR2D_Net_Archeo, improves smooth inversion results and buried archeological remains' size and position can be delineated from those enhanced models.    KEYWORDS: DC Resistivity, 2D, Inversion, Deep Learning, archaeo-geophysics.   ACKNOWLEDGEMENT: This study is part of the Ph.D. thesis of the first author and the manuscript about this study has been submitted to Pure and Applied Geophysics. This study is also made under the Ankara University Technopolis R&D projects (STBP code: 084286).