Cottonseed germination prediction with non-destructive NIR strategies

This study investigates the use of near-infrared (NIR) spectroscopy combined with chemometric modelling to non-destructively predict the germination potential of individual cottonseeds ( Gossypium hirsutum , variety BRS Aroeira). A total of 400 seeds were analyzed using one benchtop model (NIR1, operating in the 1000–2500 nm range) and two portable devices — NIR2 (908–1676 nm) and NIR3 (1350–2500 nm). Germination tests were conducted according to standardized protocols, and spectral data were pre-processed using Savitzky-Golay derivatives, standard normal variate (SNV), and advanced filters, such as Generalized Least Squares Weighting (GLSW), to correct for instrumental and temporal variability. Partial Least Squares Discriminant Analysis (PLS-DA) models were developed to classify seeds as germinated (G) and non-germinated (NG). Among the tested instruments, the benchtop NIR1 delivered the highest classification accuracy (94.3%), while the portable devices NIR2 and NIR3 yielded predictive accuracies of 87.2% and 83.3%, respectively. Notably, integrating GLSW significantly improved model robustness, particularly for portable devices. The best-performing models identified spectral regions associated with key biochemical components, such as lipids, proteins, and cellulose, which underpin the physiological traits linked to germination. These findings demonstrate that portable NIR technology is a practical and rapid tool for in-field seed quality assessment, providing a valuable solution to enhance productivity and support precision agriculture strategies in cotton production. • Portable NIR spectrometers to predict individual cotton seed germination; • GLSW and EPO preprocessing methods are needed; • Optimization of data set separation in model building steps; • Potential for greater productivity gains than random cotton planting.