Abstract: The safety inspection system for large vehicles at the entrance and exit of nuclear facilities based on digital radiation imaging can conduct internal perspective inspections of vehicles, helping to discover prohibited items hidden inside the vehicles, ensuring nuclear safety, and preventing nuclear material leakage. To improve the efficiency of system detection and reduce the risk of manual image review and missed detection, data-driven radiation image intelligent recognition algorithms can be adopted. However, such algorithms require a large number of fully annotated image datasets containing prohibited items. Obtaining the above data through experiments is costly, and some controlled prohibited items cannot be physically obtained, which limits the application of intelligent detection methods. This article proposes a radiation image simulation data generation method based on model file projection for dual projection vehicle inspection systems. This method can convert the point surface model of prohibited items into a three-dimensional digital model of prohibited items containing material and structural information. Based on this, the digital model of prohibited items is simulated and projected according to the geometric parameters of the dual projection system. The image is overlaid on the actual vehicle background projection image collected by the system to obtain a pair of up and side projection perspective images containing prohibited items, forming a training dataset for intelligent detection algorithms. The results indicate that this method can generate a diverse and reliable dataset of hazardous material radiation images without the need for real physical experiments.