In order to improve the safety supervision ability of waterway transportation, and provide effective information technology support to maritime administrative law enforcement officers, a top-down and bottom-up knowledge graph construction method was proposed to solve the problems of decentralized and complex knowledge and difficult query of vessel on-site supervision in the smart maritime scene. By analyzing the triple structure of vessel onsite supervision and building an ontology model, the entity recognition and knowledge extraction were carried out by using sequence annotation based model and crawler technology, and domain knowledge was integrated by using binary classification model, and knowledge storage and visualization were carried out by using graph database Neo4j. Taking Dafeng port in Yancheng city as an example, the knowledge graph of vessel on-site supervision was constructed and experimentally verified. The results show that the construction method has higher accuracy, the knowledge graph technology can effectively correlate with the vessel on-site supervision knowledge, strongly support the maritime law enforcement personnel to backtrack and use the inspection events, and provide a new method for realizing intelligent maritime.