Traditional resistivity inversion methods tend to rely on the initial inversion model selected, get stuck in local minima, and be time-consuming. To address these issues, this study proposed a real-time resistivity inversion method based on the Res-UNet neural network. First, a significantly expanded forward response dataset was generated using the Gmsh software. Then, inversion experiments were carried out based on appropriate network parameters determined according to data characteristics. The experimental results indicate that the Res-UNet algorithm can fully dig the data characteristics and rapidly produce resistivity images that align with the electrical properties of strata. The experiments on the dataset for resistivity forward modeling yielded a mean squared error between the predicted values and the forward responses of 0.019 44, and those on the test set yielded a mean squared error of 0.075 8, suggesting improved imaging results compared to traditional inversion methods. Furthermore, the proposed method achieved encouraging results in the inversion calculations of simulation models, enabling rapid and accurate inversion of the location and morphologies of subsurface anomalies while exhibiting a strong noise resistance. This study provides a new method and philosophy for mapping the relationship between resistivity data and the actual geoelectric structures.