Research on Optimizing the Model of Graph-Image Integration in Information Visualization Using Deep Learning
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
In the field of big data-driven information visualization, the inadequate accuracy of cross-modal fusion of graphs and images is a key issue hindering decision-making efficiency in scenarios such as healthcare and finance. This paper proposes a dual-branch attention fusion optimization algorithm. This algorithm constructs a graph feature extraction branch by improving the graph convolutional neural network (GCN) (introducing node importance weight coefficients to optimize topological feature extraction). It also designs an image feature extraction branch by combining residual connections and dilated convolutions to mitigate gradient vanishing and expand the receptive field. Finally, a spatial-channel dual attention fusion module is implemented to achieve dynamic feature complementation. To validate the algorithm's performance, experiments were conducted on the MedicalVis-1.0 (5000 medical samples) and FinanceVis-2.0 (4000 financial samples) datasets, comparing three mainstream algorithms. Results show that our algorithm achieves a PSNR of 38.62dB (a 20.1% improvement over CNN-Splice) and a SSIM of 0.923 (a 9.5% improvement over GCN-CNN-Fix) on MedicalVis-1.0, with a runtime of 0.85s/frame (a 19.0% reduction over GCN-CNN-Fix). On FinanceVis-2.0, it achieves a PSNR of 37.85dB and an SSIM of 0.915, with subjective evaluation scores of 9.0 (for medical scenarios) and 8.9 (for financial scenarios), outperforming both competing algorithms and effectively improving the accuracy and efficiency of cross-modal information visualization.
Fields of Study & Subjects
Multimodal Machine Learning Applications
Advanced Image Fusion Techniques
Generative Adversarial Networks and Image Synthesis
Publication details
- Journal / source
- IEEE International Conference on Power, Electronics and Green Energy (ICPEGE)
- Pages
- 184–189
- Publication date
- 28 January 2026
- Language
- EN