Abstract: Objective Aiming at the problem of incomplete wellbore coverage and blank bands in formation microresistivity imaging caused by the distribution of instrument plates and the pushing mechanism, in order to overcome the limitations of traditional filling methods such as easy distortion and difficulty in maintaining fine structures like fractures in strongly heterogeneous layers, the AOTGAN network based on generative adversarial networks is adopted to fill the blank bands. To achieve high-precision and high-fidelity information reconstruction. Methods A high-quality dataset was constructed based on the original electrical imaging map and the CIFLog full-wellbore filling map. Adaptive context awareness and multi-scale feature enhancement mechanisms were introduced into the GAN network. Combined with four loss functions for dynamic optimization, an AOT-GAN network that takes into account both global semantics and local details was formed. Based on the image evaluation indicators, the hyperparameters were selected. This network was used to fill the electrical imaging images of different slit network shapes and texture features, and the effects were compared with those of the classic GAN network, Criminisi algorithm and Bicubic interpolation method. Results and Conclusions AOT-GAN outperforms the classical algorithms in both peak signal-to-noise ratio (32.93 dB) and structural similarity index (77.58%). The filling effect is natural and seamless, and can effectively maintain the continuity of high-angle seams and reticular seams, and accurately restore texture details such as convolute laminations and flintstone nodules. It provides reliable data support and theoretical basis for the calculation of reservoir parameters based on formation microresistivity imaging.