Construction of visual knowledge graph of crop diseases and pests based on deep learning

In response to the complex entity relationships, data aggregation difficulties, and knowledge sharing challenges in the field of crop pests and diseases, this study leverages the advantages of knowledge graph structure to propose a deep learning-based construction method. This method is grounded in domain ontology and employs a new annotation model to transform entity and relationship extraction into sequence labeling problems, with simultaneous labeling to enhance efficiency. It directly models ternary relationships to address the challenge of extracting overlapping relationships. Using the BERT-BiLSTM +CRF end-to-end model for experimentation, its F1 score reaches 91.34%, outperforming various classic models. Finally, the extracted knowledge is stored in a Neo4j graph database to achieve knowledge visualization and inference, providing a high-quality knowledge base for downstream applications.