The semantic analysis technology of image has always been a difficult point in image field. As an intelligent and efficient way of organizing, knowledge graph can help users accurately query the information. This paper firstly puts forward an image semantic analysis process based on knowledge graph, then adopts deep learning model to describe image's features. Image semantic knowledge fusion and processing are studied on this basis. A multilevel image semantic model is constructed to manage entity triples and support automatic construction. Finally, applications for semantic retrieval and visualization are analyzed which has some guiding significance for information organization and knowledge management of media semantic.