Sensor noise informed representation of hyperspectral data, with benefits for image storage and processing

Area of Science:

• Remote Sensing
• Image Processing
• Sensor Physics

Background:

• Hyperspectral image processing often overlooks sensor noise information, despite its importance.
• Current radiance data representations can lead to suboptimal results in spectral analysis and compression ratio calculations.

Purpose of the Study:

• To define alternative hyperspectral data representations that make sensor noise accessible for image processing.
• To enable more accurate noise estimation, data compression, and algorithm reformulation.

Main Methods:

• Defined a "corrected raw data" representation, proportional to photoelectron count.
• Developed a variance-stabilized representation using a square-root transformation of the photodetector signal.

Main Results:

• Corrected raw data allows simpler noise estimation and more compact storage.
• Variance-stabilized data achieves signal-independent noise, reduces data volume by nearly half, and offers a better measure of uncompressed data size.
• Both representations facilitate the incorporation of sensor noise into hyperspectral processing algorithms.

Conclusions:

• Alternative data representations are crucial for leveraging sensor noise in hyperspectral imaging.
• These new representations can lead to more robust and efficient hyperspectral data analysis and processing.