Proximate Composition Profiling and Adjustment of Near-Infrared Spectroscopy (NIRS) Equations for Dry Legumes from Local Markets in Northern Thailand and Southern China

Proximate analysis is essential for assessing the nutritional composition of legumes. Near-Infrared Spectroscopy (NIRS) is widely used as a rapid, non-destructive method. However, commercial NIRS calibration models are primarily developed using major commercial legumes and may not represent the spectral variability of underutilised legumes, limiting their transferability across species. This study aimed to evaluate the proximate composition of local market legumes and adjust existing NIR equations for predicting proximate constituents. Twenty-six dry seed legume samples, representing four genera and multiple species from northern Thailand and southern China, were analysed using standard wet-chemistry methods. Reference analyses showed that soybeans contained the highest protein and fat levels among the legumes studied. Partial Least Squares regression was used to adjust prediction models, with 5–7 newly acquired samples per model. Model performance was evaluated using leave-two-out cross-validation. The adjusted models yielded root mean square error values of 0.987, 0.586, 0.671, 0.363, and 0.976 for protein, fat, moisture, ash, and fibre, respectively. Spectral adjustment improved agreement between predicted and reference values, particularly for protein, fat, and ash. This study demonstrated the initial potential of adapting existing NIRS calibration models to diverse legumes. Future work must prioritise independent external validation with expanded sample cohorts to establish operational robustness.