Removal of Artifacts in Borehole Images Using Machine Learning

Abstract

In this paper, a supervised machine-learning (ML) method to remove artifacts and noise from borehole images is described. Borehole images may exhibit a variety of issues and artifacts due to reasons such as environmental and thermal noise, imperfect calibration, and current leakage through the tool body. Methods that are currently employed to improve these images are based on traditional signal-processing techniques. Although these methods are capable of removing the artifacts in images and significantly improving image quality, they have some drawbacks as well. These drawbacks include not being entirely suitable for real-time implementation and issues with reproducibility. The alternative method presented here is based on an ML algorithm that is trained using a data set pairing raw data with data processed using a traditional signal-processing-based approach. The resulting ML model is capable of being implemented in near-real time. Furthermore, the application of the algorithm does not require user supervision, increasing the reproducibility of the results.