光纤光谱技术对猕猴桃品质及成熟度的无损检测

Nondestructive detection for kiwifruit quality and maturity by optical fiber spectroscopy technology

猕猴桃可溶性固形物含量(SSC)和硬度是评价其品质的关键参数,同时也是判别其成熟度的重要指标。为探究基于光纤光谱技术预测猕猴桃SSC、硬度和成熟度的可行性并寻求最佳预测模型。首先,采用光纤光谱(200~1000 nm)采集系统获取不同成熟期“贵长”猕猴桃的反射光谱,并测定SSC和硬度的参考值。接着,基于全光谱和参考值构建偏最小二乘回归(PLSR)和主成分回归(PCR)预测模型。然后,应用连续投影算法(SPA)和竞争性自适应重加权算法(CARS)选取特征波长,构建简化的多元线性回归(MLR)和误差反向传播(BP)网络预测模型。最后,通过偏最小二乘判别分析(PLS-DA)和简化的K近邻(SKNN)算法,构建预测猕猴桃成熟度检测模型。结果表明:CARS-BP模型对SSC的预测性能最优,其预测集决定系数R_(P)^(2)=0.90,预测集均方根误差(RMSEP)和剩余预测偏差(RPD)分别为0.64和3.22;CARS-MLR对硬度的预测性能相对最优,其R_(P)^(2)=0.83,RMSEP和RPD分别为1.67和2.47;PLS-DA模型对猕猴桃成熟度的检测性能最优,其正确识别率高达100%。该研究为水果品质和成熟度的无损检测提供重要指导。

The soluble solids content(SSC)and firmness of kiwifruit are two important indices for evaluating its quality and distinguishing its maturity.In this paper,we explore the feasibility of predicting the SSC,firmness,and maturity of kiwifruit using optical fiber spectroscopy technology and of finding the best prediction model.First,an optical fiber spectroscopy(200~1000 nm)acquisition system was used to collect the reflectance spectra of the different maturity stages of‘Guichang’kiwifruit.Simultaneously,the reference values of the SSC and firmness were measured.Two methods,namely,partial least-squares regression(PLSR)and principal components regression(PCR),were employed to establish the models on the basis of the full spectra and the reference values.Then,multiple linear regression(MLR)and an error back-propagation(BP)network were applied to build simplified models on the basis of the selected characteristic variables from the full wavelengths using the methods of successive projection algorithm(SPA)and competitive adaptive reweighted sampling(CARS).Finally,partial least-squares discrimination analysis(PLS-DA)and a simplified K nearest neighbor(SKNN)algorithm were applied to build models for predicting the maturity of kiwifruit.The results showed that,for the SSC,the CARS-BP model had the best prediction ability(R_(P)^(2)=0.90,RMSEP=0.64,RPD=3.22),and for the firmness,the CARS-MLR model had the best prediction ability(R_(P)^(2)=0.83,RMSEP=1.67,RPD=2.47).The PLSDA model had the best detection ability,and the maturity discrimination accuracy was up to 100%.These results can provide important guidance for the nondestructive prediction of the quality and maturity of fruits.