Recent Advancements in Hyperspectral Image Reconstruction from a Compressive Measurement

Area of Science:

• Computational Imaging
• Deep Learning
• Spectral Imaging

Background:

• Hyperspectral (HS) image reconstruction is crucial for recovering high-resolution spectral data from compressive measurements.
• Deep neural networks have significantly enhanced HS reconstruction accuracy and efficiency.

Purpose of the Study:

• To provide a comprehensive overview of recent progress in HS image reconstruction.
• To systematically categorize and analyze existing reconstruction paradigms and their advancements.

Main Methods:

• Categorization into traditional model-based, deep learning-based, and hybrid frameworks.
• Examination of sparsity/low-rank priors, CNNs to Transformers, and deep unfolding strategies.
• Review of benchmark datasets, evaluation metrics, and prevailing challenges.

Main Results:

• Deep learning approaches, including Transformers, show significant improvements over traditional methods.
• Hybrid models effectively integrate data-driven priors with mathematical modeling.
• Key challenges include spectral distortion, computational cost, and generalizability.

Conclusions:

• The field has advanced significantly due to deep learning integration.
• Addressing current challenges is vital for future progress in HS image reconstruction.
• This survey serves as a reference for researchers and practitioners.