Electrical Resistivity Tomography Image Enhancement using Autoencoder Neural Network with Synthetic Data Validation

Abstract

The development of neural network is gradually improving in the recent years, it shows great performance in data pattern learning for several applications. A type of the neural network called autoencoder has successful in image processing such as noise reduction or anomaly detection, this can also be applied to the indiscernible images of various geophysical prospecting survey. The resistivity tomography images from direct-current resistivity (DCR) method are vague requiring specialists' analysis to delineate the appropriated models of subsurface. The ability of autoencoder in image transformation is suitable to clarify these images. The developed neural network has intention in enhancing the inverted resistivity model to gain the precision in interpretation process. The pairs of synthetic models and the solution of resistivity distribution are computed by reliable software to train the neural network to be able to transform the ambiguous resistivity images to unambiguous ones. The results show that the autoencoder neural network can resolve the unclear problem of the synthetic resistivity distribution from DCR method of certain kinds of structure according to the training set of the data.