基于近红外光谱与高斯过程的高粱单宁含量快速检测  

Rapid Detection of Sorghum Tannin Content Based on Near-Infrared Spectroscopy and Gaussian Process

作　　者：赵瑾熠 陈争光[1] 衣淑娟[2] ZHAO Jin-Yi;CHEN Zheng-Guang;YI Shu-Juan(College of Information and Electrical Engineering,Heilongjiang Bayi Agricultural University,Daqing 163000,China;Engineering College,Heilongjiang Bayi Agricultural University,Daqing 163000,China)

机构地区：[1]黑龙江八一农垦大学,信息与电气工程学院,大庆163000 [2]黑龙江八一农垦大学,工程学院,大庆163000

出　　处：《分析化学》2024年第7期1028-1038,共11页Chinese Journal of Analytical Chemistry

基　　金：国家自然科学基金项目(No.52275246);黑龙江八一农垦大学“三纵”支持项目(No.ZRCPY202214)资助。

摘　　要：高粱是我国主要的酿酒原料之一,也是重要的饲料原料.在酿酒过程中,高粱籽粒中的单宁含量对酒品品质具有决定性作用;作为饲料原料时,高粱的单宁含量对饲料的利用率有重要影响,因此高粱中的单宁含量对其品质和用途具有重要影响.传统方法检测高粱中单宁含量时存在耗时长和成本高等问题,本研究利用近红外光谱结合化学计量学方法实现了高粱单宁含量的快速无损检测.在对光谱进行预处理的基础上,使用蒙特卡洛交叉验证法(MCCV)结合高斯过程回归(GPR)进行异常样本剔除;然后,将样本集随机划分为建模集和预测集,使用无信息变量消除法(UVE)进行特征波长选择;最后,建立GPR模型,并与偏最小二乘回归(PLSR)模型和支持向量机回归(SVR)模型进行性能对比.结果表明,GPR模型的性能全面优于PLSR和SVR模型,经去趋势组合Savitzky-Golay卷积平滑进行预处理,剔除异常样本并进行特征波长选择后,建立的GPR模型为最优模型,其建模集决定系数(R^(2)_(C))、预测集决定系数(R^(2)_(P))和相对分析误差(RPD)分别为0.9979、0.9529和4.8453.本研究结果表明,采用近红外光谱结合化学计量学方法建立的GPR模型可用于高粱单宁含量的快速无损检测.The tannin content of sorghum seeds had a significant impact on the wine's quality during the brewing process.Additionally,when used as a feed ingredient,the tannin content had a major impact on feed consumption.Thus the tannin content of sorghum has a substantial impact on its quality and application.To quickly and nondestructively determine the tannin content of sorghum,near-infrared spectroscopy was combined with chemometrics in this study,which eliminated the need for time-consuming and costly conventional approaches.Following the spectra's preprocessing,anomalous samples were removed by using a combination of Gaussian process regression(GPR)and Monte Carlo cross-validation(MCCV).The sample set was then randomly divided into a modeling set and a prediction set,with feature wavelength selection carried out using the elimination of uninformative variables(UVE)method.Subsequently,a GPR model was developed,and its performance was compared with partial least squares regression(PLSR)and support vector machine regression(SVR)models.The results indicated that the GPR model outperformed the PLSR and SVR models in all aspects.The optimized GPR model,generated following pre-processing process such as Detrending and Savitzky-Golay smoothing,elimination of anomalous samples,and selection of feature wavelengths,demonstrated superior performance,with model set determination coefficient(R_(c)^(2)),prediction set determination coefficient(R_(P)^(2)),and relative percent deviation(RPD)values of 0.9979,0.9529,and 4.8453,respectively.These findings validated the effectiveness of the GPR regression model,which integrated near-infrared spectroscopy with chemometrics,for the rapid and non-destructive detection of sorghum tannins.

关 键 词：近红外光谱 化学计量学 高粱单宁 高斯过程法 快速无损检测 

分 类 号：O657.33[理学—分析化学] TS210.7[理学—化学]