Often, the ground penetrating radar (GPR) is not considered a precise and rigorous measurement tool, but rather a tool for sub-surface exploration. One way of getting precise information about sub-surface objects and their surrounding media is excavation, which is usually demanding, time-consuming, and sometimes unfeasible. For that reason, the GPR has been widely used in numerous fields. However, the correct interpretation of acquired recordings tends to be challenging especially when the subsurface dielectric constant is unknown. To avoid wrong interpretations, the precise estimation of the dielectric constant is necessary. Underground cylindric objects appear within the B-scans as hyperbola-shaped features with different characteristics. A widely used approach for dielectric constant estimation is the hyperbola fitting method. Still, numerous factors can have an impact on the features and thus reduce the effectiveness of such a method. Accordingly, with the assistance of the genetic programming technique, the dielectric constant estimations of the hyperbola fitting method can be improved. The performance of the methods is evaluated by utilizing 5-fold cross-validation along with MAE, RMSE, and R2 measures. Results reveal that cylindrical object diameter has a significant impact on the dielectric constant estimation of the hyperbola fitting method. Therefore, the proposed integration of the hyperbola fitting method with genetic programming successfully captures the impact of the diameter and improves dielectric constant estimation in GPR B-scans.