VIS–SWIR hyperspectral imaging for campaign-year verification and moisture screening in smoked paprika (Capsicum annuum L.)

Smoked paprika from three production campaigns (2023–2025) was analysed by line-scan hyperspectral imaging (HSI) to assess campaign-year discrimination and screening of reference indices. Ninety-one commercial lots were scanned with two complementary cameras, Specim FX10 (400–1000 nm) and Specim FX17 (900–1700 nm). Spectra were preprocessed using scatter-correction and derivative-based methods, and principal component analysis was first used to identify informative preprocessing strategies. Supervised models were then developed under repeated nested cross-validation using partial least squares discriminant analysis (PLS-DA), support vector machines (SVMs), and partial least squares regression (PLSR). Both cameras achieved reproducible campaign-year discrimination, with balanced accuracies of 96–98%, and linear models performed comparably to radial basis function SVMs. Variable-importance profiles highlighted carotenoid-related bands in the visible region and O–H/C–H overtone regions in the SWIR. Reference measurements showed between-campaign variability in American Spice Trade Association (ASTA) colour indices, whereas moisture followed a clearer storage-related trend. Moisture prediction from FX17 spectra showed the strongest regression performance (r = 0.926; RMSE = 1.138), while ASTA D0, ASTA D3 and colour loss showed lower predictive performance. Overall, VIS–SWIR HSI provided a robust framework for campaign-year discrimination and rapid moisture screening in smoked paprika. • VIS–SWIR HSI discriminates smoked paprika production campaigns. • Repeated nested cross-validation supports robust 96–98% balanced accuracy. • Linear models matched non-linear kernels in parsimonious classification. • SWIR-based moisture prediction supports rapid screening for lot segregation. • ASTA colour varies by campaign without a strictly monotonic ageing pattern.