FMI Logs Deblurring and Inpainting Using Deep Learning

Micro-resistivity image logs have been utilized widely for studying the geological features of drilled formations. These image logs contain significant information required for their correct interpretation. Unfortunately, due to the design of the Formation Micro Imager (FMI) tool, a considerable portion of data is missing from the final image log and is visualized as blank stripes. Missing data can severely threaten the correct interpretation of the FMI logs. In addition, some other artifacts, like image blur appearing because of the complex operational conditions, may significantly affect the data quality, complicating the interpretation process. In this work, the Deep Image Prior (DIP) deep learning technique was utilized for the inpainting of FMI logs. Furthermore, the DIP was applied to inpainting several outcrop images to mimic the FMI log image restoration. Then, the restored outcrop images were compared to the original to assess the inpainting goodness and answer the question of how close the restored FMI logs might be to the “reality.” Finally, DIP with regularization by denoising (RED) was utilized to demonstrate the example of FMI log deblurring. The study has shown that applying both techniques may be a viable approach for restoring corrupted FMI logs completely.