Abstract: To reveal the droplet breakup mechanisms induced by mixing vane spacer grids during the reflood phase of a Large-Break Loss-of-Coolant Accident (LBLOCA) in a Pressurized Water Reactor (PWR), visualization experiments on single droplets impacting an inclined surface under film boiling conditions were conducted, specifically targeting the structural characteristics of mixing vanes. By incorporating the SAM2 deep learning model and the Weka machine learning algorithm, a hybrid image processing method adapted to severe droplet deformation and micro-droplet splashing was developed, enabling the precise extraction of droplet dynamic parameters. The results indicate that under high Weber number (We) film boiling conditions, the unique "central jet" phenomenon induced by the mixing vanes leads to a distinct "double-group distribution" characteristic in downstream droplets: a group of extremely small droplets generated by edge shearing and a group of large droplets resulting from jet rupture. With the increase of the Weber number, the number of secondary droplets exhibits exponential growth, leading to a significant expansion of the gas-liquid interfacial heat transfer area. This study confirms the limitations of the traditional single-size assumption. The constructed breakup source term and size model, which account for the double-group characteristics, provide critical support for improving the prediction accuracy of thermal-hydraulic subchannel codes.