Deep Error-Aware Iterative Optimization Network for Broadband Mosaiced Hyperspectral Imaging.

Area of Science:

• Optics and Photonics
• Computer Vision
• Signal Processing

Background:

• Snapshot hyperspectral imaging (HSI) faces challenges like low signal-to-noise ratio, limited spectral range, and poor spatial resolution.
• Existing methods struggle to effectively fuse multi-source image data for enhanced HSI reconstruction.

Purpose of the Study:

• To develop a novel hyperspectral imaging system integrating broadband mosaic and high-resolution (HR) panchromatic (PAN) images.
• To introduce a deep learning network, EIONet, for reconstructing HR HSI with improved spectral quality.

Main Methods:

• Proposed a new HSI acquisition paradigm combining broadband mosaic and HR PAN images.
• Developed the Deep Error-aware Iterative Optimization Network (EIONet) with a Hierarchical Error-aware Cube Updating Mechanism (HECUM).
• Employed a Physics-Based Spectral Degradation Modeling approach for accurate degradation process modeling.

Main Results:

• EIONet successfully reconstructs HSI with high spatial resolution and spectral quality.
• HECUM effectively suppresses error accumulation by prioritizing difficult image regions.
• Achieved state-of-the-art performance on two public datasets across multiple evaluation metrics.

Conclusions:

• The proposed system and EIONet offer a new paradigm for HR HSI acquisition and reconstruction.
• EIONet demonstrates significant improvements in spatial and spectral fidelity for hyperspectral imaging.
• The method provides a robust solution for overcoming limitations in current HSI technologies.