LCTC: Lightweight Convolutional Thresholding Sparse Coding Network Prior for Compressive Hyperspectral Imaging.

Area of Science:

• Optics and Photonics
• Computer Vision
• Signal Processing

Background:

• Compressive spectral imaging enhances spatial and spectral data.
• Current methods use hand-crafted priors for sparsity, often leading to suboptimal results.
• Identifying the correct sparse transform domain is challenging.

Purpose of the Study:

• To develop a novel method for adaptive sparse transform domain identification in compressive spectral imaging.
• To improve the performance and efficiency of compressive spectral imaging techniques.

Main Methods:

• Proposed a convolutional sparse coding-inspired untrained network prior.
• Designed a Lightweight Convolutional Thresholding sparse Coding (LCTC) network.
• Employed self-supervised learning for transform domain and coefficient identification.
• Integrated the LCTC prior as Plug-and-Play (PnP) regularization into optimization algorithms.

Main Results:

• The LCTC network effectively identifies sparse transform domains and coefficients.
• PnP-LCTC demonstrated superior performance compared to existing methods.
• Experiments validated the approach's effectiveness and efficiency across various scenarios.

Conclusions:

• The proposed LCTC network and PnP-LCTC offer a significant advancement in compressive spectral imaging.
• This self-supervised, adaptive approach overcomes limitations of hand-crafted priors.
• The method is effective and efficient for enhancing spatial and spectral information.