Abstract: In this study, arabica coffee cherries (cv. Catimor) from Yunnan were harvested at four maturity stages and systematically analyzed for changes in the contents of key flavor precursors (such as sugar, acid, amino acids, and fatty acids). The feasibility of using hyperspectral imaging technology (398.5–1 000.2 nm) combined with chemometrics for rapid and non-destructive maturity discrimination was explored. The results showed that the contents of soluble solids and total sugar significantly increased with maturity, while the contents of titratable acidity (TA) and total protein showed fluctuating changes. By comparing multiple spectral preprocessing methods, it was found that the standard normal variate transformation combined with a partial least squares discriminant analysis model achieved the best maturity discrimination performance (with a validation set accuracy of 95.26%). The partial least squares regression model had the best prediction effect for total sugar content (R2v = 0.96), but exhibited a lower prediction accuracy for the TA content, which varied highly across maturity stages. This study provides an effective technical approach for the precise harvesting and non-destructive quality detection of fresh coffee cherries.

Key words: Yunnan arabica coffee; maturity; hyperspectral; non-destructive testing; flavor precursors