Rapid, Non-Destructive Prediction of Starch and Moisture Content in Saffron Corms Using Visible–Near-Infrared Spectroscopy Combined with Machine Learning

The starch and moisture content of saffron corms are critical indicators of their flowering potential and yield. This study investigated the use of rapid, non-destructive vis-NIR reflectance spectroscopy measurement to assess these parameters. The measurements were used to develop predictive models through four machine learning algorithms (PLSR, RF, SVR, and GPR). Spectral data were obtained from 130 fresh corm samples. Wavelength analysis identified key starch-sensitive intervals (~930–1000 nm and ~1150–1220 nm) and a broad moisture-sensitive region (~900–1350 nm). Among the evaluated models, the partial least squares regression (PLSR) model demonstrated the optimal predictive performance for moisture (R² = 0.89, RMSE = 0.91%, RPD = 3.67, RPIQ = 4.91) and moderate performance for starch (R² = 0.68, RMSE = 26.29 mg g⁻¹, RPD = 1.87, RPIQ = 2.37, dry weight). These results demonstrated the viability of VIS-NIR spectroscopy as a viable, non-destructive tool for the pre-planting assessment of saffron corm quality. The method provides a practical foundation for corm screening and selection, with potential for further improvement in starch prediction through expanded calibrations and advanced modeling techniques.