.png){kind=icon}
.png){kind=icon}
Return
Grayscale-driven snapshot hyperspectral imaging via efficient phase modulation
.png){kind=icon}
PRE
AI
DOI：10.1016/j.optcom.2025.132459.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
It has always been treated as challenging work to obtain hyperspectral imaging with a compact, high-speed, and high-resolution solution for dynamic scene capture. Existing snapshot hyperspectral imaging methods predominantly rely on coded multi-channel RGB reconstruction, limiting the spectral compression efficiency and applicability beyond the visible spectrum. To overcome this limitation, we have proposed a phase-encoded snapshot hyperspectral imaging method using only coded single-channel grayscale image. By employing a liquid crystal on silicon spatial light modulator, the proposed method implements an efficient spatial-spectral encoding and compression mechanism. As an enhanced deep learning framework AD-ResUNet has been introduced within the reconstruction process, the channel recovery ratio of single-channel reconstruction can be improved to three times that of multi-channel reconstruction. Experimental results demonstrate that compared with the multi-channel scheme, our method achieves PSNR up to 34.144 dB and SSIM up to 0.946 across 400-700 nm. This study validates the feasibility of coded grayscale based hyperspectral reconstruction and advances snapshot hyperspectral imaging through the enhancements of system architecture, channel recovery ratio optimization, and algorithmic improvements, providing a viable pathway toward practical implementation.
Journal
IF:
2.5
Papers:
595
Citations:
2.7W