Cross-variety validation of a polarimetric glare correction algorithm for maize

Remote phenotyping in maize is hindered by glare, contaminating spectral data. This study expands on prior work using a Mueller matrix bidirectional reflectance distribution function (mmBRDF) model and correction network (CN) to mitigate this. We tested the mmBRDF/CN’s applicability across three maize varieties (B73, MO17, and a Pioneer Hybrid) using a multistatic fiber-based instrument. Leaf sample measurements, model parameter analysis, and Monte Carlo simulations were used to quantify CN performance. Refractive index (n) was consistent. Additionally, vairance in σ may aid in varietal differentiation. The CN significantly reduced error and variance in GNDVI across all varieties, with RMSE decreasing from approximately 0.10 to 0.035, and standard deviation decreasing by a factor of 2. As expected, outlier analysis showed low sun angles and specular viewing caused most residual error; yet, generally, the CN had broad applicability for correcting glare in the tested genotypes.