Mitigating Illumination-, Leaf-, and View-Angle Dependencies in Hyperspectral Imaging Using Polarimetry

Area of Science:

• Agricultural Science
• Remote Sensing
• Optics

Background:

• Plant phenotyping advances crop yield and breeding through automated monitoring.
• Sun glare from leaves in field imaging creates noise, hindering accurate data collection.
• Polarization data can differentiate surface reflection from internal leaf scattering.

Purpose of the Study:

• To develop a facile algorithm using polarization data to decouple leaf surface glare from internal scattering.
• To improve the accuracy of plant phenotyping data acquired by imaging sensors in field conditions.

Main Methods:

• Combined data from a mast-mounted hyperspectral imaging polarimeter (HIP) and a fiber-based Mueller matrix bidirectional reflectance distribution function (mmBRDF) instrument.
• Fitted mmBRDF data to a model to obtain parameters for simulation.
• Trained a shallow neural network using simulated data to correct HIP sensor data based on vegetation indices and polarized light.

Main Results:

• The developed algorithm significantly reduced errors and standard deviations in vegetation index calculations (GNDVI and red-edge reflection ratio).
• An improvement of an order of magnitude or more in mean error (ϵ) was observed.
• A reduction of 1.5 to 2.7 in standard deviation (ϵ) was achieved after applying the correction network.

Conclusions:

• Polarization-based algorithms can effectively mitigate sun glare artifacts in hyperspectral imaging of crops.
• This approach enhances the reliability of automated plant phenotyping for agricultural research.
• Improved data quality supports more accurate crop health monitoring and accelerates breeding programs.