A multi-objective optimization method based on a neural network surrogate model is proposed for the shape optimization of low-altitude airships. The Isight and Fluent software packages are integrated to conduct design of experiments, generating sample points for constructing the neural network surrogate model. With the minimization of aerodynamic drag and envelope surface area as the optimization objectives, the NSGA-II algorithm is employed to perform the optimization based on this surrogate model. The results show a 20.9% reduction in aerodynamic drag and a 10.9% decrease in envelope surface area. Furthermore, the influence of the weight ratio in the composite objective function and the wind speed on the optimization outcomes is investigated through parameter analysis, which involves adjusting the drag-to-surface-area weight ratio and varying the inlet wind speed. The results indicate that under different weight ratios and wind speed conditions, the improvement rates for both drag and surface area remain relatively stable, further validating the effectiveness of the proposed method across various design scenarios. This approach not only enhances the aerodynamic performance and structural efficiency of the airship but also significantly improves optimization design efficiency, providing a valuable reference for the shape design of low-altitude airships.