为准确识别酱香型高温大曲类型，对30个出仓曲检测了中红外光谱，在主成分分析中，3类大曲（黑曲、黄曲、白曲）呈现出各自聚类的趋势；进一步建立了偏最小二乘偏最小二乘判别（PLS-DA）模式识别方法，模型R2Y为0.956，Q2为0.906，可有效判别不同质量大曲类型，为生产投料配比提供数据依据。为快速定量高温大曲中类黑素，对发酵过程高温大曲建立了基于近红外光谱技术的类黑素定量模型，以60个样品建立模型，光谱经多元散射校正（MSC）结合一阶导数处理，在10 000~4 000 cm-1范围，主成分数为8时，偏最小二乘（PLS）模型效果最优，校正集决定系数R2Cal为0.987 7，校正均方根误差（RMSEC）为0.169 6，验证集决定系数R2Val为0.900 7，交叉验证均方根误差（RMSECV）为0.491 1；以15个样品做外部预测以验证模型可靠性，预测均方根误差（RMSEP）为0.460 6，标准偏差与预测标准偏差比值（RPD）为2.63，且与参考方法之间无显著性差异（P=0.772），可较好地预测未知大曲中类黑素含量。该方法操作简便，检测分析时间仅为10~15 min，效率比传统方法提高至少8倍。

To accurately identify the three different types of high-temperature sauce-flavor Daqu, 30 samples were collected from storage and subjected to mid-infrared spectra. The three Daqu categories (black, yellow, and white) clustered separately in the principal component analysis. Furthermore, a pattern recognition model was established based on midinfrared spectroscopy combed with partial least squares discriminant analysis (PLS-DA). With an R2Y of 0.956 and a Q2 of 0.906, the model effectively distinguished the different qualities and types of Daqu, offering a data-driven basis for feeding the materials during production. A near-infrared spectroscopy based quantitative model involving 60 samples during different fermentation processes was established to rapidly quantitate melanoidins in Daqu. The obtained spectrum was processed by multiplicative scatter correction (MSC) and first-order derivatives. The PLS model achieved optimal results in the range of 10 000~4 000 cm-1 and when the principal component was 8. The coefficient of determination for the calibration set (R2Cal) was 0.987 7, root mean square error of calibration (RMSEC) was 0.169 6, coefficient of determination for the validation set (R2Val) was 0.900 7, and cross-validation root mean square error (RMSECV) was 0.491 1. An external prediction with 15 samples was conducted to validate the reliability of the model, yeilding a root mean square error of prediction (RMSEP) of 0.460 6. The ratio of the standard deviation to the prediction standard deviation (RPD) was 2.63. Furthermore, there is no significant differences between the near-infrared method and the reference method (P=0.772). Therefore, this model can effectively predict melanoidin content in unknown Daqu samples. This method could be applied to the rapid quality evaluation of Daqu due to its convenience, with a detection time of only 10~15 min and an efficiency that is at least eight times higher than the traditional method.

贵州省工信厅发展专项资金科技创新项目（202209）；贵州省科技成果应用及产业化计划项目（黔科合成果[2020]2Y045）；遵义市科技计划项目（遵市科合R&D[2020]31号；遵市科合支撑GY（2021）40号）