Abstract: Traditional radar data processing algorithms face challenges such as difficulty in parameter tuning in complex environments and high costs for upgrading and evolution when targets and environments change. In recent years, radar data processing algorithms based on artificial intelligence technologies like long short-term memory and YOLO have significantly improved adaptability in complex environments. However, these methods are limited by the large workload of manually extracting features and insufficient generalization across different systems and scenarios, hindering their promotion and large-scale application in system. Inspired by large language models, a "pre-training-fine-tuning" paradigm for intelligent radar data processing is proposed and the "Houyi" foundation model based on simulation and real-measured data is constructed in this paper. The goal is to revolutionize the research and development mode of radar data processing, establish a pipeline for intelligent algorithm development and deployment, and achieve efficient evolution and large-scale application of radar intelligent data processing. The "Houyi" foundation model is based on the Transformer architecture and focuses on three primary tasks: false alarm suppression, association, and filtering. By constructing massive training datasets from simulation and real-measured data across multiple systems and scenarios, the pre-trained foundation model features strong generalization and robustness. Highly-efficient data processing for multiple systems in various operating scenarios can be achieved by fine-tuning the foundation model.