Coronal Mass Ejection (CME) is a significant phenomenon triggering severe space weather events. Visualizing the dynamic propagation of CMEs aids in deepening the understanding of their physical mechanisms and supports early warning systems. Augmented Reality (AR) technology, by integrating virtual and real environments, offers intuitive visualization of CME propagation characteristics. However, existing image feature extraction and matching algorithms face challenges such as insufficient feature representation and poor performance under noise interference, limiting AR’s effective application in space weather research. To address these issues, this paper proposes a CME image matching algorithm based on a Dual-Domain Attention Enhanced Fusion Network (DDFN). This method fuses the local perception ability of Convolutional Neural Networks (CNN) with the global modeling strength of Vision Transformers (ViT), introducing spatial and channel attention mechanisms to enhance multi-dimensional feature representation. Additionally, it integrates the Simple Contrastive Learning of Representations (SimCLR) framework to improve robustness against noise, illumination changes, and other complex conditions. Experimental results demonstrate that the DDFN achieves Top-1 matching accuracies of 81.76% and 68.44% on original and noise-added datasets, respectively, outperforming current deep learning models. Furthermore, the algorithm is successfully deployed within an AR system, enabling efficient image matching and dynamic visualization of CME propagation. This advancement significantly improves the dynamic visualization level and key feature capture in CME propagation research. Future work will focus on optimizing the model architecture and expanding practical applications to further enhance the accuracy and applicability of space weather disaster early warning technologies.
DownLoad: