Fig. 1. SSVF model schematic

Fig. 2. Architecture of composite graph convolution module

Fig. 3. Schematic of multiscale mix convolution module

Fig. 4. Architecture of VAFormer

Fig. 5. Spatial distribution diagram of pixels in CAVE、Harvard dataset

Fig. 6. Spectral information distribution of CAVE、Harvard dataset. (a) h1 image in Harvard dataset; (b) oil_painting image in CAVE dataset

Fig. 7. Heat map of correlation distribution between different spectral bands of CAVE、Harvard dataset

Fig. 8. Continuous density estimation between different spectral bands for CAVE dataset flower_ms

Fig. 9. Heatmap of spatial correlation between different patches of CAVE、Harvard dataset

Fig. 10. Comparison of absolute error plots for local zoom in the 10th channel of the CAVE data cloth_ms. (a) Grand truth; (b) CSTF; (c) UAL; (d) TSFN; (e) PZnet; (f) FF-former; (g) LGAR; (h) SSVF

Fig. 11. Comparison of absolute error plots for local zoom in the 10th channel of the Harvard data imga3. (a) Grand truth; (b) CSTF; (c) UAL; (d) TSFN; (e) PZnet; (f) FF-former; (g) LGAR; (h) SSVF

Fig. 12. Comparison of image element spectra at different positions of beads_ms for the CAVE dataset. (a) (160, 200); (b) (400, 400); (c) (200, 160)

Fig. 13. Comparison of image element spectra at different positions of imgh7 for the Harvard dataset. (a) (400, 500); (b) (400, 20); (c) (10, 10)