拉曼光谱与中红外光谱融合技术快速定量食用酒精的乙醇浓度

Integration technology of Raman and mid-infrared spectroscopy for rapid quantification of ethanol concentration in edible alcohol

中文关键词:  拉曼光谱  中红外光谱  光谱数据融合  乙醇溶液

英文关键词:Raman spectroscopy  mid-infrared spectroscopy  spectrum integration technology  ethanol solution

中文摘要:

      目的 利用拉曼光谱与中红外光谱的数据融合技术实现对食用酒精乙醇浓度(酒精度)的快速定量检测。方法 首先, 分别采集不同浓度食用酒精水溶液的拉曼光谱与中红外光谱。其次, 采用多元散射校正(multiplicative scatter correction, MSC)、卷积平滑(Savitzky-Golay, S-G)、一阶求导的方法对原始数据进行预处理。然后, 基于自举软缩减法(bootstrapping soft shrinkage, BOSS)和无信息变量消除算法(uninformative variable elimination, UVE)分别对预处理后的光谱数据进行特征提取, 并利用X-Y距离样本集划分法(sample set partitioning based on joint X-Y distance, SPXY)将光谱数据划分为校正集和预测集。最后, 建立基于拉曼光谱－中红外光谱数据融合的偏最小二乘回归(partial least squares regression, PLSR)食用酒精乙醇浓度预测模型, 并利用麻雀搜寻算法优化的混合核极限学习机算法(sparrow search algorithm-optimized hybrid kernel extreme learning machine, SSA-HKELM)提升预测性能, 实现对不同浓度食用酒精的快速、准确定量检测。结果 与拉曼光谱数据、中红外光谱数据以及中红外与拉曼光谱的数据层融合构建的预测模型相比, 中红外光谱与拉曼光谱特征层融合数据构建的预测模型具有更好的预测性能。其中, 最优模型的校正集均方根误差(root mean squared error of calibration set, RMSEC)为0.98314, 校正集决定系数(Rc2)为0.99634, 预测集均方根误差(root mean squared error of prediction set, RMSEP)为1.03256, 预测集决定系数(Rp2)为0.99036。结论 中红外光谱与拉曼光谱特征层融合预测模型可以实现对不同浓度食用酒精的高效定量检测, 为食用酒精的质量检测提供了有效的理论支持与技术保障。

英文摘要:

      Objective To achieve rapid quantitative detection of ethanol concentration (alcohol content) in edible alcohol by data fusion techniques in Raman and mid-infrared spectroscopy. Methods First, Raman and mid-infrared spectra of edible alcohol aqueous solutions with different concentrations were separately acquired. Subsequently, the original data were preprocessed using multiplicative scatter correction (MSC), Savitzky-Golay (S-G) convolution smoothing, and first-order derivative methods. Then, based on bootstrapping soft shrinkage (BOSS) and uninformative variable elimination (UVE), feature extraction was performed on the preprocessed spectral data. The spectral data were further partitioned into calibration and prediction sets using sample set partitioning based on joint X-Y distance (SPXY). Finally, a predictive model for ethanol concentration in edible alcohol was established by integrating Raman and mid-infrared spectra through partial least squares regression (PLSR). The sparrow search algorithm-optimized hybrid kernel extreme learning machine (SSA-HKELM) was employed to enhance predictive performance, enabling rapid and accurate quantification of ethanol concentration in edible alcohol at various concentrations. Results Compared to the predictive models constructed by fusing Raman spectral data, mid-infrared spectral data, and layered data from mid-infrared and Raman spectra, the predictive model based on the fusion of mid-infrared and Raman spectral feature layers demonstrated superior predictive performance. The optimal model exhibited a root mean squared error of calibration set (RMSEC) of 0.98314, a coefficient of determination of the calibration set (Rc2) of 0.99634, a root mean squared error of prediction set (RMSEP) of 1.03256, and a coefficient of determination of the prediction set (Rp2) of 0.99036. Conclusion The fusion predictive model of mid-infrared and Raman spectral feature layers enables efficient quantitative detection of different concentrations of edible alcohol, providing effective theoretical support and technological assurance for the quality assessment of edible alcohol.