SVIFNN: Robust Inpainting Fourier Neural Network for SST Scientific Visualization Image Leveraging Significant Stability and Nonsignificant Anomalies

The scientific visualization images (SVIs) of sea surface temperature (SST) play a pivotal role as visual resources for investigating oceanographic processes. However, they are often plagued by extensive data gaps due to objective factors like cloud cover. Additionally, their content deviates significantly from ordinary images (OIs), posing challenges for conventional completion techniques. Given the intricate nature of marine systems, completing the visualization of SST presents several challenges. First, predicting missing segments relies not only on prominent patterns but also on subtle anomalies, which are often overlooked by methods focused on extracting prominent features. Second, these images exhibit chaos and lack clear semantics, making it difficult for methods primarily focused on semantic extraction to effectively complete them. To address these challenges, this study presents a novel method named the inpainting Fourier neural network for SVI (SVIFNN). This approach employs a twin-stream architecture to highlight both significant stability and nonsignificant anomalies. Notably, it incorporates a “reverse attention mechanism” in the nonsignificant anomalies extraction stream (AES) to preserve unconventional information. Furthermore, by cascading Fourier neural operator (FNO), it leverages frequency-domain characteristics to mitigate spatial chaos. Through a frequency-domain feature extraction module, it achieves an adaptive fusion of significant stability and nonsignificant anomalies. The experimental results demonstrate SVIFNN’s superiority over state-of-the-art (SOTA) methods, particularly under a 68% missing rate condition. Significant improvements are observed in $R^{2}$ (18.1%, 19.3%, and 21.8%) and reductions in root-mean-square error (RMSE) (22.6%, 28.8%, and 23.8%) across different noise-to-signal (N/S) ratios of 0.1, 0.2, and 0.3, respectively, underscoring SVIFNN’s robustness in handling SVI’s extensive data gaps. Adequate ablation experiments further validate the effectiveness of the proposed nonsignificant AES and frequency-domain operators, with the latter demonstrating superior performance for SVIs compared to traditional spatial-domain convolutional neural network (CNN) and Vision Transformer (ViT) operators.