基于可见/近红外光谱技术的便携分析仪的应用

Application of a portably analyzer based on visual/near infrared spectroscopy

中文关键词:  可见/近红外光谱  无损检测  便携分析仪  苹果  可溶性固形物

英文关键词:visual/near infrared spectroscopy  non-destructive detection  portable analyzer  apple  soluble solids content

中文摘要:

      目的 为解决水果内部品质信息的快速无损检测, 自主研制了一台基于可见/近红外光谱技术的便携式分析仪, 通过试验验证其可行性及所建模型的鲁棒性。方法 以红富士苹果为检测对象, 采集透射光谱曲线, 与化学指标可溶性固形物含量(soluble solid content, SSC)分别建立基于平均光谱、基于各采样光谱的偏最小二乘(partial least squares, PLS)回归模型, 比较预测精度并对非同批次样本进行预测。结果 试验表明该分析仪对苹果SSC具有较高的测量精度, 特别是基于各采样光谱的PLS模型, 对同批次样本预测相关系数(Rp)达到0.924, 预测均方根误差低至0.429%Brix, 预测精密度(平均偏差)低至0.136%Brix, 对非同批次样本SSC表现出较强的鲁棒性能, 预测均方根误差为0.531%Brix。结论 通过此项研究, 表明该便携分析仪可用于水果内部品质信息的定量分析, 并建议采用基于各采样光谱建立的回归模型用于外来样本的预测。

英文摘要:

      Objective To solve the fast and non-destructive testing of the quality information of fruits, a portable analyzer was independently developed based on visual/near infrared spectroscopy, to validate the feasibility of this analyzer and the robustness of potential model in practical application. Method In this experiment, ‘Fuji’ apples were chosen, and partial least square (PLS) regression model was comparatively built based on the averaged spectrum and every sampling spectrum respectively, and between the transmitted spectra and chemical soluble solids content (SSC). The precision and accuracy of models were compared and validated by external batch samples. Result Results showed that this fruit analyzer had a good ability of measurement for apples’ SSC, especially the PLS model which was built on the basis of every sampling spectra, with correlation coefficient of prediction (Rp) of 0.924, root mean squared error of prediction (RMSEP) of 0.429%Brix, as well as the precision of prediction (averaged bais) of 0.136%Brix. Additionally, strong robustness was presented on the external batch samples with RMSEP of 0.531%Brix. Conclusion Overall, this study indicates the portable analyzer can be used for quantitative analysis of fruit internal quality, and the regression model to predict external samples is advised to be created based on every sampling spectrum.