HyperReconNet: Joint Coded Aperture Optimization and Image Reconstruction for Compressive Hyperspectral Imaging.

Area of Science:

• Optics and Photonics
• Computer Vision
• Machine Learning

Background:

• Coded aperture snapshot spectral imaging (CASSI) enables rapid 3D hyperspectral image (HSI) acquisition but often suffers from low reconstruction accuracy.
• Existing methods independently optimize coded apertures or reconstruction algorithms, limiting overall performance improvements.
• A unified framework is needed to simultaneously enhance both components for better HSI reconstruction.

Purpose of the Study:

• To develop a novel deep learning framework for improving HSI reconstruction accuracy in CASSI systems.
• To jointly optimize the coded aperture design and the HSI reconstruction process within a unified end-to-end method.
• To leverage convolutional neural networks (CNNs) for enhanced CASSI performance.

Main Methods:

• A CNN-based end-to-end framework was proposed, integrating coded aperture design and HSI reconstruction.
• A repeated pattern for the coded aperture was designed, with its elements learned as network weights.
• HSI reconstruction was performed by exploiting spatial and spectral correlations using deep learning.

Main Results:

• The proposed method achieved superior HSI reconstruction accuracy compared to state-of-the-art techniques.
• Experimental results demonstrated significant improvements in both quantitative metrics and perceived image quality.
• The unified deep learning framework effectively connected and optimized coded aperture design and reconstruction.

Conclusions:

• The proposed CNN-based unified framework significantly enhances HSI reconstruction accuracy for CASSI systems.
• Joint optimization of coded aperture and reconstruction via deep learning offers a powerful approach to overcome CASSI limitations.
• This method represents a significant advancement in snapshot hyperspectral imaging technology.