Efficiently and accurately obtaining aerodynamic performance is of great significance for the design and optimization of bridge girder cross-sections. Traditional wind tunnel tests and computational fluid dynamics (CFD) numerical simulations are time-consuming and expensive when obtaining the aerodynamic performance of bluff body sections, seriously affecting the efficiency of bridge wind resistance design. Therefore, predicting aerodynamic performance based on aerodynamic shape has become a key issue in structural wind-resistant design. Based on deep learning, this paper proposes an intelligent prediction method for aerodynamic performance that is not limited to the specific form of the main girder cross-section. This method is not restricted by the specific form of the cross-section. By combining Unet and ResBlock residual modules, it establishes a nonlinear mapping target from aerodynamic shape to aerodynamic performance and has strong versatility. Specifically, it transforms the traditional angular coordinate form and scalar form into a field form. Taking data fields similar to pictures as input, it is not limited by the specific form of the main girder cross-section (number of polygon corners, whether there are slots, etc.). At the same time, based on the one-to-one correspondence between shape and stable flow field, it skips multiple steps such as local sampling and pressure integration. It uses the form of a stable flow field similar to pictures to describe the aerodynamic performance of the main girder boundary and the surrounding space. Taking the stable flow field of CFD numerical simulation as output. Error analysis with CFD calculation results shows that this intelligent prediction method can effectively predict structural static coefficients and surface pressure distributions, and realizes an exponential reduction in time cost. It is expected to become a key means for bridge design and optimization.