Rapid quality evaluation of Pu-erh raw tea storage years via hyperspectral imaging coupled with machine learning and microbial profiling.

This study developed a non-destructive approach for identifying storage years and quality evolution of Pu-erh raw tea by integrating sensory evaluation, chemical analysis, electronic nose/tongue, hyperspectral imaging, and 16S rRNA sequencing. Results revealed that Bacillus and Pantoea shifts drive chemical transformations; notably, Bacillus abundance reached 40.44% in Hangzhou-stored samples after six years, correlating with significant degradation of polyphenols and ester-type catechins. Specific wavelengths at 1633, 1085, and 1250 nm were intrinsically linked to alterations in caffeine, polyphenols, and amino acids, respectively. Among tested models, Random Forest coupled with Standard Normal Variate preprocessing achieved 100% classification accuracy in the near-infrared band. Furthermore, Random Forest regression successfully visualized bacterial succession trends based on spectral data. Overall, this integrated strategy provides a rapid, non-invasive methodological framework for tea quality control, offering a valuable reference for the future expansion and validation of diverse tea aging processes.