Edge-aware adaptive hyperspectral acquisition via synergistic dual-camera architecture

Area of Science:

• Optics and Photonics
• Computer Vision
• Signal Processing

Background:

• Snapshot hyperspectral imaging captures spatial and spectral data in one exposure.
• Current methods reconstruct data cubes from sparse spectral measurements, often causing artifacts due to insufficient sampling.

Purpose of the Study:

• To develop an adaptive hyperspectral acquisition system to overcome limitations of insufficient spectral sampling.
• To enhance scene characterization and high-fidelity data reconstruction in snapshot hyperspectral imaging.

Main Methods:

• An adaptive hyperspectral acquisition system combining an edge-aware spectral sampling strategy and a deep reconstruction network.
• The sampling strategy adaptively allocates measurement resources based on scene structure.
• A reconstruction network utilizes 3D convolutional kernels for spatial-spectral dual filtering.

Main Results:

• The proposed system demonstrates reduced reconstruction artifacts and improved data quality.
• Adaptive sampling effectively allocates resources for better scene sensing.
• Deep reconstruction leverages spatial-spectral correlations for high-fidelity restoration.

Conclusions:

• The integrated approach of adaptive sampling and deep reconstruction offers a robust and efficient solution for snapshot hyperspectral imaging.
• This method enhances the performance and reliability of high-speed hyperspectral applications.