基于多成熟度光谱信息融合的阿森泰克苹果品质预测模型研究

Prediction Model of Aztec Apples Quality Based on the Fusion of Multi-maturity Spectral Information

不同成熟度的阿森泰克苹果品质变化大,会显著影响采后贮藏与销售效益。本研究以江苏宿迁四个成熟阶段的阿森泰克苹果为研究对象,首先利用主成分分析(principal component analysis,PCA)和线性判别分析(linear discriminant analysis,LDA)分析其色泽(L^(*)、a^(*)、b^(*)值)、硬度(firmness,FI)、可溶性固形物(soluble solid content,SSC)、可滴定酸(titratable acidity,TA)、水分含量(moisture content,MC)、干物质(dry matter content,DMC)的变化规律;同时,基于可见-近红外(visible and near-infrared,Vis-NIR)与近红外(nearinfrared,NIR)光谱技术结合连续投影(successiveprojectionsalgorithm,SPA)、竞争性自适应重加权(competitive adaptive reweighted sampling,CARS)、无信息变量消除(uninformative variable elimination,UVE)算法进行相关特征变量筛选,基于偏最小二乘(partial least squares,PLS)与支持向量机(support vector machine,SVM)建立阿森泰克苹果品质预测模型。结果表明,SSC、a^(*)、L^(*)、b^(*)对不同成熟度阿森泰克苹果的聚类贡献率较高,510~680、1170~1270、2300 nm为高相关度特征波段。SPA-PLS、SPA-SVM模型能很好地预测不同成熟度阿森泰克的L^(*)、b^(*)、a^(*)值,相对预测偏差(relative percent deviation,RPD)均高于3.00,CARS-PLS模型可以很好地预测SSC,RPD为3.19,但FI、TA、MC、DMC的SPA-PLS模型预测精度相对较低,RPD分别为2.27、2.21、2.32、2.42。研究结果证明Vis-NIR和NIR光谱方法能够预测不同成熟度阿森泰克苹果品质,为阿森泰克苹果采收管理与质量安全控制提供技术参考。

The quality of Aztec apples varies significantly at different maturity stages,which can have a significant impact on postharvest storage and sales efficiency.This study focused on Aztec apples at four different maturity stages in Suqian,Jiangsu Province.Firstly,the variations in color(L^(*),a^(*),b^(*)values),firmness(FI),soluble solid content(SSC),titratable acidity(TA),moisture content(MC)and dry matter content(DMC)were analyzed using principal component analysis(PCA)and linear discriminant analysis(LDA).Simultaneously,visible and near-infrared(Vis-NIR)and near-infrared(NIR)spectral techniques,along with the successive projections algorithm(SPA),competitive adaptive reweighted sampling(CARS)and uninformative variable elimination(UVE)algorithms were employed for selecting relevant characteristic variables.Subsequently,partial least squares(PLS)and support vector machine(SVM)were utilized to establish quality prediction models for Aztec apples.The results revealed that SSC,a^(*),L^(*)and b^(*)had a significant impact on the categorization of Aztec apples at different maturity stages.Notably,wavelength bands in the ranges of 510 to 680 nm,1170 to 1270 nm and 2300 nm exhibited strong correlations with characteristic attributes.The SPA-PLS and SPA-SVM models demonstrated remarkable performance in predicting the L^(*),b^(*)and a^(*)values of Aztec apples at different maturity stages,with all relative percent deviation(RPD)values exceeding 3.00.The CARS-PLS model effectively predicted SSC with an RPD of 3.19.However,the prediction accuracy of SPA-PLS models for FI,TA,MC and DMC was comparatively lower,with RPD values of 2.27,2.21,2.32 and 2.42,respectively.The results demonstrated that Vis-NIR and NIR spectroscopy methods could predict the quality of Aztec apples at different maturity stages,providing valuable technical references for the harvest management and quality control of Aztec apples.