Cho G.-H., Kim Y.-J., Jeon K., Kang K.-S., 2025. Development of acorn discrimination model for warm-temperature evergreen oaks using hyperspectral analysis. Ann. For. Res. 68(1): 125-136. https://doi.org/10.15287/afr.2025.4012
Abstract
We used hyperspectral analysis to distinguish between acorns of Japanese red oak (Quercus acuta Thunb.) and ring-cup oak (Quercus glauca Thunb.), two closely related species of the evergreen oaks. To accomplish this, 631 Japanese red oak acorns and 505 ring-cupped oak acorns were collected from the seed orchard in Jeju Island, Korea, and hyperspectral imaging was performed. Two types of hyperspectral devices, Corning and Korea Spectral Products (KSP), were used to calibrate images and extract regions of interest. Average spectra were obtained from the extracted regions of interest, and morphological variables were added to the Corning data to form a dataset. Partial least square (PLS) was used as the learning model, Standard normal variate, Multiplicative scatter correction, and Savitzky-Golay filtering were applied as preprocessing techniques, and competitive adaptive reweighted sampling and successive projection algorithm were applied as variable selection techniques ; and the combination of preprocessing method, the number of PLS components, and the number of selected variables were optimized.. The lightweight model was generated from the selected variables, and the performance was improved by combining the morphological variables. As a result, the lightweight model based on Corning dataset showed 45~85% accuracy, and the lightweight model based on the KSP dataset showed 75~90% accuracy. The model utilizing morphological variables in the Corning-based lightweight model showed a high accuracy of 98-100%, so we were able to discriminate the acorns of evergreen oaks between Q. acuta and Q. glauca. The results of this study are expected to serve as a basis for future model development for seed classification of hybrid oak acorns.