Abstract: The operations at railway logistics yards have become increasingly heavy and complex, posing severe challenges to their safe operation. The traditional safety management mode relying on manual inspections and fixed monitoring suffered from numerous blind spots, low efficiency and delayed emergency response. This study sorted out five core safety risk factors within yards, including personnel, vehicles, cargo, equipment and environment, constructed a system architecture for intelligent safety risk identification of such yards based on air-ground collaboration, and investigated key technologies such as multimodal perception fusion, intelligent safety risk recognition and risk situation inference. The air-ground collaboration paradigm effectively overcomes the coverage blind spots and information fragmentation of single perception methods, forms a technical closed loop from "perception" to "cognition" and further to "prediction", and provides technical support for all-round, all-weather and intelligent prevention and control of safety risks in railway logistics yards.