Objective Formation microresistivity imaging (FMI) images features incomplete borehole coverage and thus contain blank strips due to the gaps between electrode pads, as well as their pushing and attachment mechanisms. Traditional blank strip filling methods suffer from several limitations, including susceptibility to image distortion and difficulty in retaining information on fine-scale structures such as fractures. This study developed an aggregated contextual-transformation generative adversarial network (AOT-GAN) for blank strip filling, aiming to achieve high-precision and high-fidelity information reconstruction.

Methods Initially, a high-quality dataset was prepared based on original FMI images, along with full-borehole images obtained through image inpainting using CIFLog. Then, adaptive context awareness and multi-scale feature enhancement mechanisms were incorporated into the GAN. Finally, through dynamic optimization using four loss functions, the AOT-GAN integrating both global semantics and local details was developed. After optimal hyperparameters were selected based on image evaluation metrics, the AOT-GAN was employed to the perform blank strip filling for FMI images with varying fracture network patterns and texture characteristics. Furthermore, the filling effects were compared to those derived using three classical algorithms, i.e., GAN, Criminisi algorithm, and Bicubic interpolation. Results and Conclusions The AOT-GAN outperformed the three classical algorithms in both peak signal-to-noise ratio (PSNR, 32.93 dB) and structural similarity index measure (SSIM, 77.58%). This network generated natural, seamless FMI images, effectively maintaining the continuity of high-angle and reticular fractures while also accurately restoring the details of textures such as convolute beddings and flint nodules. The results of this study provide reliable data support and a theoretical basis for calculating reservoir parameters based on FMI images.