Self-supervised constrained super-resolution fast coded spectral imaging system.

Area of Science:

• Optics
• Image Processing
• Spectroscopy

Background:

• Current coded aperture spectral imaging methods face limitations in speed and image quality, particularly at low sampling rates.
• Developing faster and higher-quality spectral imaging techniques is crucial for various scientific applications.

Purpose of the Study:

• To propose a novel self-supervised constrained super-resolution fast coded spectral imaging system.
• To enhance the resolution of hyperspectral images by fusing low-resolution hyperspectral images (LR-HSI) with high-resolution multispectral images (HR-MSI).

Main Methods:

• Construction of a discrete cosine transform spectrometer (DCTS) to acquire LR-HSI and HR-MSI.
• Design of a self-supervised spectral image super-resolution network (SSAM-Unet) tailored to the physical imaging process.
• Fusion of LR-HSI and HR-MSI using the SSAM-Unet to reconstruct high-resolution hyperspectral images (HR-HSI).

Main Results:

• The SSAM-Unet successfully reconstructs HR-HSI by effectively fusing LR-HSI and HR-MSI.
• The proposed system demonstrates good imaging performance and generalization ability across various experimental conditions.
• The method achieves satisfactory imaging results even at significantly low sampling rates.

Conclusions:

• The developed self-supervised constrained super-resolution fast coded spectral imaging system offers a significant advancement over existing techniques.
• This method effectively addresses the challenges of slow imaging speeds and poor image quality in spectral imaging.
• The system shows promise for applications requiring high-resolution hyperspectral data acquisition under resource-constrained sampling conditions.