Enhancing DC Resistivity Data Two-Dimensional Inversion Result by Using U-Net Based Deep Learning- Algorithm: Examples from Archaegeophysical Surveys

In this study, we suggested using a convolutional neural network (CNN) based algorithm to enhance two-dimensional (2D) Direct Current Resistivity (DCR) data inversion results. We developed U-net based CNN algorithm, named DCR2D_Net_Archaeo. We generated 1080 sets of 2D resistivity models that simulate buried archaeological remains. We calculated synthetic data for those models for different electrode arrays. We added 2% random noise to apparent resistivity data sets and inverted those data sets. We used the 2D inversion results as input and the corresponding real resistivity model as output. By using those 1080 input and output data sets we developed the DCR2D_Net_Archaeo algorithm. First, we tested this algorithm by using synthetic data. We showed that the developed algorithm improved the 2D classical smoothing regularization inversion and the buried body's geometry and depth can be found very close to the real model. Afterward, we also tested the developed algorithm with real data collected from two different archaeological sites. We showed that the buried wall cross-section location and depth are better found by the DCR2D_Net_Archaeo algorithm than by using the conventional inversion codes for the inversion of DCR data if we compare it with the excavated wall structure.