食品与发酵工业2026，Vol.52Issue(11)：339-345,7.DOI:10.13995/j.cnki.11-1802/ts.044417
基于贝叶斯算法优化Transformer结合双向门控循环单元对青稞酒的快速识别
Rapid identification of Qingke liquor based on Bayesian algorithm-optimized transformer combined with bidirectional gated recurrent unit
摘要
Abstract
In response to the phenomenon of adulteration in the market of liquor,the search for rapid,non-destructive and scientific methods for identifying the quality of liquor has become one of the research hotspots in the relevant fields.This study aims to quickly iden-tify the"Chinese Huzhu Qingke Liquor"(CHQL),a product protected by geographical indication in our country.Three types of liquor samples,namely CHQL,other brand Qingke Liquor(OBQL),and non-Qingke based Baijiu(NQBB),were collected.The spectral char-acteristics of these three types of liquor were studied using UV-near-infrared fusion spectroscopy.An attempt was made to establish a rapid identification method for CHQL.Firstly,the UV spectra and near-infrared spectra of the samples were collected,and the spectra were fused at the data level.Then,four preprocessing methods(standard normal transformation,Savitzky-Golay smoothing,first-order deriva-tive,and second-order derivative)and three classification models(BiGRU,Bayes-BiGRU,Bayes-Transformer-BiGRU)were respectively examined for the identification effect of CHQL.The results show that the data obtained after second-order derivative preprocessing com-bined with Bayes-Transformer-BiGRU has the best classification recognition effect,with classification accuracy of 94.12％,response time of 492.89 seconds,and PAM of 0.87.This indicates that Bayes-Transformer-BiGRU combined with UV-NIR fusion spectroscopy can quickly,non-destructively and accurately distinguish CHQL.关键词
"互助"青稞酒/双向门控循环单元/Transformer/贝叶斯优化Key words
Chinese Huzhu Qingke liquor/bidirectional gated recurrent unit(BiGRU)/Transformer/bayesian optimization引用本文复制引用
许诗咏,张世芝,石婷,张明锦..基于贝叶斯算法优化Transformer结合双向门控循环单元对青稞酒的快速识别[J].食品与发酵工业,2026,52(11):339-345,7.基金项目
国家自然科学基金项目(22363010) （22363010）
青海省自然科学资金项目(2022-ZJ-769) （2022-ZJ-769）
