Fig. 1. Process of snapshot compression spectral imaging of superstructure surface

Fig. 2. Binary pattern meta-atom design process. (a) 5×5 square with 70 nm side length is rotated three times to form complete metasurface unit of 10×10 pixel C4 symmetric structure; (b) squint image of meta-atom

Fig. 3. Structural parameters and optical performance of meta-atom. (a) Effect of number of metasurface units on spectral resolution and spectral fidelity; (b) transmission spectra of C4-structured meta-atom under X-polarized and Y-polarized incident light; (c) two types of meta-atom structures and their corresponding transmission curves

Fig. 4. Solution process of WCA-DUN

Fig. 5. Architecture of WCA-DUN

Fig. 6. Linear projection module and prior denoising module

Fig. 7. Diagram of windowed channel attention [including normalization layer (LayerNorm), WCA module, and feedforward network (FFN)]

Fig. 8. Fast Fourier transform (FFT) module and partial scenes from CAVE and KAIST datasets. (a) FFT module; (b) 12 scenes from CAVE and KAIST datasets

Fig. 9. Images of six spectral channels (total 28 channels) reconstructed using four methods and comparison of recovered spectra (shown as insets ① and ② on RGB images) and spatial details of selected regions

Fig. 10. Real scenes captured by hyperspectral camera and reconstruction results