Research on Image-Text Recognition Method Based on Semantic Graph Modeling

Image-text recognition faces challenges in real-time processing, semantic alignment, and adaptability to dynamic environments. To address these issues, this paper proposes a big data recognition method based on semantic graph modeling. The core innovation lies in constructing dynamic semantic networks with real-time adjustment mechanisms for deep visual-textual fusion. The method integrates CNN-based visual feature extraction and keyword-driven semantic parsing to generate cross-modal links, which are structured into a graph. This graph is dynamically optimized using attention-based correlation updates and spectral clustering. A real-time perception module maintains semantic consistency, while decision and format adaptation modules enhance recognition accuracy and robustness. Experiments show the method significantly improves matching accuracy, processing speed, and adaptability under large-scale, real-time data.