Near-infrared hyperspectral imaging at 935-1720 nm was combined with chemometrics and machine learning to predict total soluble solids, titratable acidity and their ratio in intact sweet tamarind. Optimized SVM regression models achieved prediction correlations of 0.959, 0.961 and 0.957, respectively. An SVM classifier separated fruit by the commercial acidity threshold with 82.86% accuracy. The results support nondestructive sorting, pending broader seasonal and cultivar validation.
Key findings
- Prediction correlations for TSS, TA and TSS/TA were about 0.96. • Commercial acidity classification reached 82.86% accuracy. • The intact-fruit method is compatible with sorting-line use.
Why this matters globally
The approach could reduce destructive sampling and improve grading consistency, but throughput, cost and robustness on real production lines require further evaluation.
Thai researcher contribution
Khon Kaen University, KMITL and Udon Thani Rajabhat University jointly develop technology for an economically important Thai fruit.
Limitations to consider
The abstract omits fruit count, origin, season and external testing; performance may shift across cultivars, moisture conditions and instruments.