Conditional Flow Matching for Uncertainty Aware Borehole Image Inpainting
Journal of Petroleum Geology · 10.07.2026
DOI: 10.1111/jpg.70093 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Petroleum Geology, online erschienen am 10.07.2026. Verfasst von George Bittar und 2 weiteren Autor:innen. Die Arbeit nennt 32 Literaturangaben.
Abstract (Originalsprache)
Borehole microresistivity imagers produce high‐resolution images of the borehole wall, but incomplete tool coverage leaves vertical gaps that increase interpretation uncertainty. We present a fast, uncertainty aware inpainting approach based on conditional flow matching. The model operates on a two‐channel FMI representation and jointly inpaints both the static and dynamic normalized images within a single conditional generative flow. By fixing the observed measurements and repeating inference from different random noise initializations inside the missing region, we generate an ensemble of plausible completions and estimate pixelwise uncertainty from ensemble variability. Experiments on open access FMI data show coherent cross gap continuations and smooth gap boundaries in both channels, with inference speed suitable for real time and offline interpretation workflows.
Zitieren
APA
Bittar, G., Wu, X., & Chen, J. (2026). Conditional Flow Matching for Uncertainty Aware Borehole Image Inpainting. Journal of Petroleum Geology. https://doi.org/10.1111/jpg.70093
Vancouver
Bittar G, Wu X, Chen J. Conditional Flow Matching for Uncertainty Aware Borehole Image Inpainting. Journal of Petroleum Geology. 2026. doi:10.1111/jpg.70093
BibTeX
@article{Bittar2026,
author = {Bittar, George and Wu, Xuqing and Chen, Jiefu},
title = {Conditional Flow Matching for Uncertainty Aware Borehole Image Inpainting},
journal = {Journal of Petroleum Geology},
year = {2026},
doi = {10.1111/jpg.70093}
}
EUVIMED