摘要
[目的]本文利用可见/近红外光谱定量检测山西省不同产区晋虞1号桃的可溶性固形物(Soluble solids content,SSC)含量,旨在建立一个简单有效、适应性能好的校正模型为后续在线检测设备的开发与利用提供模型参考。[方法]采集3个产区桃的可见/近红外漫反射光谱,选择不同的预处理方法消除客观因素对原始光谱的影响,比较发现SG平滑+多元散射校正(multiplicative scatter correction,MSC)预处理方法建模结果最优。采用Kennard-Stone算法以3 ∶1比例划分样品集,其中校正集270个用于建立PLS模型,预测集90个用于评价模型性能。为了简化模型运算量、提高模型预测性能使用蒙特卡罗无信息变量消除(Monte Carlo uninformative variables elimination,MC-UVE)与连续投影算法(Successive projection algorithm,SPA)相结合筛选有效特征波长。最后,比较了偏最小二乘(Partial least squares,PLS)算法所建单一产地和混合产地下晋虞1号桃SSC含量可见/近红外光谱模型的预测能力。[结果]与单一产地和两两混合产地模型相比,混合3产地桃校正集样本建立的模型预测效果最好,预测的相关系数(Rp)和预测的均方根误差(RMSEP)分别为0.949和0.652 °Brix。[结论]利用多个产地的晋虞1号桃样本建立的混合模型具有较强的包容性,可提高对晋虞1号桃SSC含量的预测精度,减小产地差异对SSC含量可见/近红外光谱检测的影响。本文可为山西省内晋虞1号桃内部品质SSC含量的无损检测模型提供了理论基础。
[Objectives] The soluble solids content (SSC) of Jinyu No.1 peach in Shanxi Province was determined by visible / near infrared(Vis/NIR)spectroscopy in order to establish a simple and effective model with good adaptability,which could provide a model reference for the development and utilization of the on-line detection equipment in the future.[Methods] Vis/NIR diffuse reflectance spectra of Jinyu No.1 peach from three different production areas were obtained,and different pre-treatment methods were selected to eliminate the influence of objective factors on the original spectrum.The most effective pre-treatment method was determined as Savitzky-Golaysmoothing combined with multivariate scattering correction(MSC)method.The samples were divided at 3 ∶1 scale by Kennard-Stone algorithm,among which 270 samples were used as a calibration set to establish PLS model and 90 samples were used as a prediction set to assess the performance of the model.In order to simplify the computation and to improve the prediction of the model,Monte Carlo information-free variable elimination(MC-UVE) combined with continuous projection algorithm(SPA)were used to filter the effective feature wavelengths.The SSC prediction ability of developed models from a single or mixed production areas were compared between each other based on the partial least square (PLS)algorithm.[Result] Compared to the models developed with samples from a single or two mixed production areas,the one developed with the mixed calibration sets from three production areas produced the best prediction ability with the correlation coefficient of prediction(Rp)of 0.949 and the root mean square error(RMSEP)of prediction at 0.652°Brix.[Conclusion]The results suggested that the model based on mixed samples from three production areas possessed a higher tolerance,and could improve the SSC prediction accuracy and reduce the influence of the different producing areas on SSC values when using visible/near infrared spectroscopy.It provided a theoretical basis for SSC nondestructive testing model to determine the inner quality of Jinyu No.1 peach in Shanxi Province.
作者
赵旭婷
张淑娟
孙海霞
邢书海
李成吉
陈彩虹
高庭耀
Zhao Xuting;Zhang Shujuan;Sun Haixia;Xing Shuhai;Li Chengji;Chen Caihong;Gao Tingyao(College of Engineering,Shanxi Agricultural University,Taigu 030801,China)
出处
《山西农业大学学报(自然科学版)》
CAS
北大核心
2019年第5期106-112,共7页
Journal of Shanxi Agricultural University(Natural Science Edition)
基金
国家自然科学基金(31271973)
晋中市科技重点研发计划(Y172007-4)