Abstract: This paper designed a Large Language Model (LLM)-based railway intelligent translation system to address the difficulties encountered by foreign passengers in cross-border railway travel, including communication barriers, inadequate multimodal interaction, and inaccurate translation of railway-specific terminologies by general translation services. The system established a cross-modal interaction framework integrating voice, text and image data, and adopted a split-type model scheme to adapt to complex railway travel scenarios. It adopted core technologies including vision LLM-enabled scenario-based recognition, non-autoregressive speech recognition and customized adaptation of textual LLMs, and efficiently and accurately fulfilled translation demands in high-frequency scenarios such as voice consultation, ticket purchase and alteration, as well as passenger travel guidance. Its translations comply with standard railway industry expressions and deliver convenient and efficient intelligent travel services. When deployed in cross-border railway travel service scenarios, the system ensures smooth communication for foreign passengers throughout their trips.