Rapid non-destructive determination of fatty acids in perilla germplasm using near-infrared hyperspectral imaging and chemometric modeling

Perilla ( Perilla frutescens L.) seeds are a rich source of functional fatty acids, including α-linolenic acid, and serve as a major plant-based omega-3 resource; however, to the best of our knowledge, non-destructive prediction of fatty acid contents using hyperspectral imaging (HSI) has not been reported. In this study, a rapid and non-destructive method was developed for predicting fatty acid contents in perilla seeds by combining near-infrared HSI with regression modeling. Hyperspectral images covering the 1,050–1,600 nm range were acquired from 236 germplasm accessions, and four spectral preprocessing methods were applied. Effective wavelengths were selected using a genetic algorithm and competitive adaptive reweighted sampling. Four regression models were evaluated, among which support vector regression yielded the optimal performance for all six components. The optimal models achieved R² p = 0.894 with a normalized root mean square error (NRMSE) of 5.650% for total fat and R² p = 0.865 with an NRMSE of 6.845% for α-linolenic acid, and five of the six components exhibited NRMSE values < 8%, with the exception of stearic acid. SHapley Additive exPlanations and principal component analysis identified key spectral regions contributing to fatty acid prediction, and non-destructive prediction at the individual seed level was demonstrated. To our knowledge, this study presents the first near-infrared-HSI-based prediction of fatty acid contents in perilla seeds, and the proposed approach is expected to serve as a high-throughput non-destructive screening tool for high-fatty-acid lines in breeding programs.