Application Research of Pattern Recognition of Fusion Knowledge Graph in Complex Scenarios

Abstract Knowledge graphs serve as an effective mode of organizing and managing knowledge in various fields, such as retrieval, updating, and question and answer. As a result, research on their construction method has gained significant attention. This paper is about pattern recognition. To deal with multimodal features, we combine multimodal data sources and use the structure of synergistic attention mechanisms, which are made up of self-attention mechanisms and guided attention mechanisms. The improved multimodal bilinear method is used for the fusion of modal data. Then, a cross-domain knowledge graph cross-embedding method is proposed to perform multi-semantic interactions for all entities and relationships in multiple domains to achieve cross-domain knowledge graph embedding. Finally, we explore the performance of the constructed knowledge graph fusion model by applying it to the problem of plant disease detection in complex scenarios. When we add 10%, 20%, and 30% noise to the image data captured in complex scenarios, the results demonstrate high detection accuracy and robustness, respectively.