Predicting the composition of aroma components in Baijiu using hyperspectral imaging combined with a replication allocation strategy-enhanced stacked ensemble learning model

Ester and acid aroma compounds are crucial components affecting the fragrance of Baijiu, and their composition can endow the Baijiu with a fruity, acidic, floral, or roasted aroma. This study aims to quantitatively detect the ester and acid content in Soy Sauce-Aroma Type Baijiu (SSAB) using hyperspectral imaging (HSI) technology and a stacked ensemble learning (SEL) model. To mitigate the impact of data imbalance, an improved oversampling technique known as the replication allocation strategy (RAS) was utilized. After comparing the study results, it was found that the established RF-RAS-SEL model yielded the best performance, with an Rp2 of 0.9803 and RMSEP of 0.3314 mg/L for predicting ester content and an Rp2 of 0.9914 and an RMSEP of 0.4565 mg/L for predicting acid content. These findings demonstrate that HSI can achieve the non-destructive and accurate detection of esters and acids in SSAB, providing a novel method for analyzing Baijiu aroma.