【作者】 李明；

【导师】 韩东海；

【作者基本信息】 中国农业大学 ， 食品科学， 2018， 博士

【摘要】 水果,作为一种重要的食物,其品质也越来越受重视。利用可见/近红外技术进行水果品质的检测研究历经多年,论文发表可谓无数,实际应用已成旧闻。然而,因水果的多样性、复杂性及特殊性,仍有一些问题亟待解决。例如,用一个指标模型预测不同品种水果的该类指标,提高模型的适应性,进而实现模型优化;科学合理地进行数据运算处理,简化模型的变量维数;引入新的优化思路,实现水果品质评价的客观性与全面性;通过一些举措,提升模型的准确性等。为此,本文开展了如下研究:(1)针对水果样品的特殊性,并结合梨样本的实验数据,提出可见/近红外光谱技术对水果品质进行预测时,在样品信息的获取、样品信息的关联以及样品信息的预测三个方面的应对措施,为后续研究奠定理论基础。(2)通用定量模型的建立与研究。选取了三种差异较大的甜瓜(玛瑙、金红宝、西州蜜)样本,建立了一个多品种可溶性固形物通用定量预测模型。通过光谱预处理之后,考虑到光谱一致性、光谱对应的化学信息以及光谱的穿透能力,预选取750 nm～950 nm这一波段,建立三种甜瓜的可溶性固形物通用定量预测模型,得出甜瓜的瓜顶部位更适合用于预测其可溶性固形物含量。通过竞争性自适应重加权算法(competitiveadaptivereweightedsampling,CARS)方法筛选变量,其筛选变量仅为24个,且大部分筛选的变量均有一定的物理及化学意义,同时有较大的回归系数。运用筛选的变量建立的模型结果表明模型准确性及可靠性均有一定程度的提高。(3)定性预测模型的变量简化与研究。使用可见/近红外光谱技术对样品进行漫透射模式获得光谱,并结合PLS-LDA模型,对“富士”苹果内部果肉是否褐变进行进行无损伤判别。通过CARS算法筛选出0.34%的变量能有效对苹果内部褐变进行判别,且正确率达到100%,而通过积分面积(peakarea,PA)对苹果内部褐变判别模型更为简单,其准确率虽不及CARS-PLS-LDA模型,但相比全波长模型而言,PA-PLS-LDA模型的准确率也有所提高,并能达到一定要求。(4)综合因子指数模型的建立与研究。运用可见/近红外光谱技术对低温藏下的“黑宝石”李果实内部信息进行无尤损检测。通过相关系数法选取各个指标的有效波段进行建模,并通过回归系数曲线揭示指标之间的关系。通过因子分析得出各个指标所占权重,提出运用综合因子指数对“黑宝石”李进行品质评价的方法。综合因子指数PLS模型,不仅可以初步预测“黑宝石”李果实在低温贮藏条件下的各个指标,也可以预测“黑宝石”李的生理状态,探讨了综合因子指数预测模型建立的可行性。(5)模型的颜色补偿优化方法研究。以“黑宝石”李果实在低温贮藏期间多个指标发生变化这一现象着手,通过多种数据处理方法揭示指标间的相互关系,并提出可见/近红外光谱水果无损检测模型的影响因素。结果显示,除可溶性固形物之外,“黑宝石”李果实在低温贮藏期间的其它指标间均有一定相互关系,且每一种指标与三个颜色指标(L*、a*、b*)的相关性程度各有不同。通过二维可见/近红外相关光谱分析得出,颜色的变化与果肉指标的变化是相互影响的,而且颜色变化的速度是要快于果肉其它化学以及物理指标的变化。运用初级数据融合技术建立果肉颜色补偿PLS模型,通过模型结果得出,除可溶性固形物之外,其它指标果肉颜色补偿PLS模型的准确度均有一定程度提高,且提高了所建模型的解释能力。更多还原

【Abstract】 Fruit,as an important food source,is more and more important.Visible and near infrared spectroscopy(Vis-NIR)technology has been widely used by testing fruit quality because of its advantages of lossless,fast and environment-friendly.At the same time,the paper has been published for many years and the actual application has become old news.However,some problems still need to be solved due to the diversity,complexity and particularity of fruit.For example,a model is used to predict the quality of different varieties or similar fruits to optimize the model’s adaptability;the model’s variable dimension is simplified through scientific and reasonable processing of data;the introduction of new optimization thinking reflects the objectivity and comprehensiveness of fruit quality evaluation;the accuracy of the model is improved through some measures.The specific results are as follows:(1).On the basis of the particularity of fruit samples and combing with data of pear samples,three aspects were puted forward the operation details.These three aspects are sample information collection,information correlation and information prediction.The above laid the theoretical foundation for the further researches.(2).Development and research of general quantitative model.Three different varieties of melons(Manao,Jinhongbao and Xizhoumi)were selected to study the generality of fruit quantitative prediction model.After the spectral pretreatments,the spectral consistency,the corresponding chemical information and the penetrating capacity of the spectrum were considered.A general model of SSC of three melon was established in the preselected 750 nm～950 nm band,and it was concluded that the stylar-end of melon was more suitable for evaluating its quality.The 24 variables were selected by the competitive adaptive reweighted sampling(CARS)algorithm,and most of the selected variables had certain physical and chemical meanings,while there was a large regression coefficient.The results showed that the accuracy and reliability of the model were improved to some extent after variables selection.(3).Variable simplification and research of discriminant model.Using the Vis-NIR to conduct a diffuse transmission of the sample,the PLS-LDA model was used to distinguish the browning of the"Fuji" apple without damaging the sample.Through the competitive adaptive reweighted sampling(CARS)algorithm,0.34%of the variables could be selected effectively to judge the browning of apple,and the accuracy rate was 100%.It is simpler to judge the browning model by integrating peak area(PA),which is less accurate than the CARS-PLS-LDA model.Compared with the full-wavelength model,the accuracy rate of PA-PLS-LDA model was also improved,and it could meet certain requirements.(4).Development and research of comprehensive factor index model.Quality parameters in ’Friar’plums were assessed non-destructively during low temperature storage using Vis-NIR.Throughcorrelation coefficient method,the effective band of each parameter was selected to model and therelationship between parameters was revealed through regression coefficient curves.The parameters’factor weightings were calculated using factor analysis(FA)to build a comprehensive PLS model forfurther assessment.The comprehensive PLS model could not only preliminarily predict the variousparameters of the ’Friar’ plums during low temperature storage,but also could predict the physiological status of ’Friar’ plums,which realized the feasibility of establishing comprehensive prediction model.(5).Optimization and research of color compensation model.Focusing on phenomenon of several parameters changes of the ’Friar’ plums during low temperature storage period.The correlations among parameters were revealed through a variety of data processing methods,and the factors influencing the non-destructive testing model of visible/near-infrared spectra were proposed.The results showed it had a certain relationship among parameters during the low temperature storage period except SSC,and each parameter had a different degree of correlation with the three color parameters(L*,a*,b*).Through the analysis of the two dimensional visible and near-infrared correlation spectra,the change of color and the change of the flesh parameters were mutually influenced,and the color changed faster than other chemical and physical parameters of flesh.Through the primary data fusion technology,the flesh color compensation PLS model was established.According to the model results,except SSC,the flesh color compensation PLS model of other parameters had a certain degree to improve the accuracy of the model.更多还原