InSPECtor: an end-to-end design framework for compressive pixelated hyperspectral instruments

Area of Science:

• Optics and Photonics
• Computer Vision
• Data Compression

Background:

• Traditional hyperspectral imaging systems acquire dense spatial and spectral data, often requiring post-acquisition compression.
• Existing methods face challenges in managing large data volumes and long acquisition times.

Purpose of the Study:

• To develop a framework for designing optimized, micropatterned snapshot hyperspectral imagers.
• To enable data compression at the sensor level for hyperspectral imaging.

Main Methods:

• Implementation of a framework using TensorFlow for joint optimization of filter array layout and reconstructor.
• Utilizing automatic differentiation for optimizing the micropatterned filter array and reconstructor design.
• Exploration of compression ratios based on filter passbands, scanning frames, and filter layouts.

Main Results:

• Demonstration of a snapshot hyperspectral imager design achieving significant data compression.
• Reduction in data volume, acquisition time, and detector space by a factor of 40 compared to dense sampling.
• Restoration of full hyperspectral data cubes from compressed sensor-level data.

Conclusions:

• The joint optimization framework enables substantial data volume reduction for hyperspectral imagers.
• Snapshot hyperspectral imagers offer a more efficient alternative to classic dense sampling methods.
• This approach facilitates efficient data acquisition and processing in hyperspectral imaging.