Abstract: To address the cumbersome and uncontrollable generation of ammunition transport plans caused by the complex coupling of multiple constraints in ammunition transport tasks, a semantic graph retriev-al-augmented visual generation system for ammunition transport planning is proposed to improve the fea-sibility, completeness, and interactive optimization capability of solution generation. First, natural lan-guage tasks are parsed into key task patterns, and semantic expansion and constraint association are con-ducted with the assistance of a large language model (LLM) to construct an ammunition transport semantic graph. Then, multi-level retrieval queries are generated based on the semantic graph structure, and highly relevant case knowledge is aggregated through multi-directional semantic matching and ranking mecha-nisms, while recommendation guidance is provided by the LLM in combination with a Sankey diagram structure. Subsequently, semantic alignment and result ranking are performed among the semantic graph, knowledge fragments, and a structured knowledge base, with the LLM assisting in recommending key combinations of transport elements. Finally, designed a visual generation system that, through multi-view linkage and LLM-driven interactive operations, enables multi-stage optimization and tactical experience feedback during solution generation. The effectiveness and practicality of the proposed system are vali-dated through case studies of aircraft carrier ammunition transport and army wartime ammunition resupply, as well as expert evaluations.