Hyperspectral Compressive Snapshot Reconstruction via Coupled Low-Rank Subspace Representation and Self-Supervised...

Area of Science:

• Optics and Photonics
• Computer Vision
• Signal Processing

Background:

• Coded aperture snapshot spectral imaging (CASSI) captures 3D hyperspectral images (HSIs) but faces reconstruction challenges.
• Existing methods struggle with implicit HSI features, require extensive training data, and offer limited interpretability.

Purpose of the Study:

• To develop a novel HSI reconstruction method that overcomes limitations of current approaches.
• To improve reconstruction accuracy, generalization performance, and interpretability of 3D HSIs.

Main Methods:

• A model-driven low-rank subspace representation decomposes HSIs.
• A data-driven deep guided spatial-attention network (DGSAN) learns deep coefficient priors (DCP).
• Self-supervised training integrates priors into an iterative optimization framework without requiring paired data.

Main Results:

• The proposed method demonstrates enhanced reconstruction accuracy for 3D HSIs.
• Improved generalization capabilities across different datasets and imaging systems were observed.
• The approach offers better interpretability compared to existing techniques.

Conclusions:

• The novel HSI reconstruction method effectively leverages spectral correlations and deep priors.
• The self-supervised, data-driven approach provides a robust solution for CASSI.
• This work advances 3D HSI reconstruction with superior performance and interpretability.