Hybrid Space Calibrated 3D Network of Diffractive Hyperspectral Optical Imaging Sensor

Area of Science:

• Optical imaging
• Multispectral imaging
• Computational imaging

Background:

• Diffractive multispectral optical imaging is crucial for optical sensing.
• Spatially variant point spread functions commonly cause image blurring in these systems.
• Existing methods may require complex equipment or extensive data.

Purpose of the Study:

• To introduce a novel, high-quality, and efficient hybrid space calibrated 3D network (HSC3D) for correcting spatially variant blur in diffractive multispectral imaging.
• To achieve high-accuracy restoration of eight-channel multispectral images.
• To improve image quality without demanding complex hardware modifications or large datasets.

Main Methods:

• Development of the HSC3D algorithm, integrating a 3D U-Net architecture with space calibration modules (magnification and rotation effects).
• Implementation of 3D convolutional layers to enhance image processing capabilities.
• Establishment of a custom diffractive multispectral imaging system with a manufactured diffractive lens and corrected refractive lenses.

Main Results:

• The HSC3D algorithm demonstrated significant improvements in image quality.
• Mean peak signal-to-noise ratio (PSNR) increased by 3.33 dB.
• Mean structural similarity index (SSIM) improved by 0.08 compared to a typical Unrolled Network algorithm.

Conclusions:

• The HSC3D algorithm effectively restores multispectral images degraded by spatially variant blur.
• The method offers superior performance and high space calibration ability, enhancing imaging quality.
• This approach presents a new, practical method for diffractive multispectral image sensing with potential applications in diffraction lens spectroscopy.