Accuracy Evaluation of Visible-Near Infrared Spectroscopy for Detecting Insect Damage in Acorns of <i>Quercus acuta</i>

Abstract This study aimed to test near-infrared spectroscopy to assess insect damage to oak acorns collected from a seed orchard of Quercus acuta in Jeju Island, Korea. A total of 550 acorns were sorted into 362 sound and 178 unsound (insect-damaged) acorns, followed by near-infrared spectroscopy. To minimize spectral data errors, preprocessing techniques such as first derivative, multiplicative scatter correction, standard normal variate, and Savitzky-Golay filter were applied, along with multivariate analysis methods like partial least squares. Then the model performance, including accuracy and precision, was evaluated using the Variable Importance in Projection. The near-infrared wavelength of the acorns showed strong absorption peaks at 660~720nm and a slight downward trend at 900~1000nm. The most effective model for distinguishing unsound acorns was Savitzky-Golay filtering treatment applied in the 400~1000nm range and used partial least squares, showing prediction accuracy of 86 % (p&lt;0.05). The performance was significantly influenced by absorption points at 660~720nm and 960~1000nm, with the latter range believed to be affected by changes in moisture and carbohydrates due to insect damage. The former range showed lower classification capability due to chlorophyll and color variation but affected the model performance when used with near-infrared wavelength range. These findings can narrow down the scope of investigation for future research using wider wavelength ranges or multispectral analysis.