摘要
为了建立稳定可靠的鲜枣品质检测模型,利用光谱和水分补偿方法进行鲜枣内部品质的检测。首先,针对鲜枣各品质指标(水分含量、可溶性固形物含量、维生素C含量、蛋白质含量、硬度值),采用回归系数法(RC)提取特征波段并建立最小二乘支持向量机(LS-SVM)检测模型,预测集的决定系数(R2P)均在0.8261以上,预测均方根误差(RMSEP)均在3.324 9以下。在提取各项品质指标特征波段的基础上,剔除其他四项单一品质特征波段中与水分特征波段(包含利用RC法所提取到的水分特征波长和鲜枣中具有明显水分特征的吸收峰)重叠或接近的波段,并与鲜枣水分含量值进行数据融合建立了各项指标的水分补偿模型。结果表明,硬度值的水分补偿模型精度有一定提高,R2P和RMSEP分别为0.830 5和0.055 3;可溶性固形物含量、维生素C含量、蛋白质含量的水分补偿模型精度均有所下降,R2P分别为0.804 1,0.878 2和0.837 8,RMSEP分别为1.347 3,0.638 0和3.503 2。然后,分析各品质指标间的相关性,结果表明,水分含量在0.05水平上与硬度值呈现显著的相关性,在0.01的水平上与其余三项品质指标之间存在极显著的相关性,相关性强弱与水分补偿模型的建模结果相互支持。研究表明,水分补偿法所建的预测模型可用于鲜枣内部品质的检测,水分含量与其他四项品质指标之间有相互作用并影响其他品质指标所建立的预测模型。该研究为进一步探讨光谱检测中各内部品质指标间交互作用的解耦提供了新思路。
In order to establish a stable and reliable detection model to identify the quality of fresh jujube,the visible/near-infrared reflection spectroscopy techniques and the method of moisture compensation were used to detect the internal quality of fresh jujube.Moisture content（MC）,soluble solid content（SSC）,firmness,soluble protein content（PC）and vitamin C content（VC）were used as internal quality index of Huping Jujube,regression coefficient（RC）was applied to select effective wavelengths and least squares-support vector machines（LS-SVM）models were built based on the effective wavelengths,respectively.The results of the five RC-LS-SVM models were obtained with the determination coefficient of every prediction（R2P）of MC,SSC,PC,VC,firmness as 0.859 5,0.884 0,0.867 1,0.909 9 and 0.826 1,respectively.The root mean square error of prediction（RMSEP）of MC,SSC,PC,VC,firmness were 1.243 1,1.005 3,3.324 9,0.479 8 and 0.056 7,respectively.Then,wavelengths overlapped with or closed to characteristic wavelengths of moisture content were removed from characteristic wavelengths of SSC,PC,VC and firmness,respectively.Characteristic wavelengths of moisture content were composed of distinct moisture absorption peak on fresh jujube（960,1 200,1 400,1 780 and 1 900 nm）and characteristic wavelengths selected by RC of PLSR model of moisture content.Characteristic wavelengths after the moisture compensation of each index（SSC,PC,VC,firmness）was used to carry out data fusion with moisture content of fresh jujube,moisture compensation LS-SVM model of each index（SSC,PC,VC,firmness）was built based on fused data,respectively.The results indicated that the model＆#39;s accuracy of firmness was improved after moisture compensation,R2P and RMSEP were 0.830 5 and 0.055 3,respectively.The results also revealed that the model accuracy of SSC,VC and PC were reduced respectively after moisture compensation,R2P were 0.804 1,0.878 2 and 0.837 8,respectively and RMSEP were 1.347 3,0.638 0 and 3.503 2 respectively.Finally,the correlation relationship between the quality indexes was analyzed.The results indicated that an significant correlation relationship was revealed between moisture content and firmness in the 0.05 level,an extremely significant correlation relationship was revealed between moisture content and any of the other three indexes（SSC,PC,VC）in the 0.01 level.This research shows that prediction model based on the method of moisture compensation can be effective to realize evaluation of the internal comprehensive quality on Fresh Jujube.What＇s more,there is an interaction between moisture content and any of the other four indexes.In fact,prediction models based on the other quality indexes are affected by moisture content.This research provides a new method for the decoupling of interaction between the various internal quality indexes in the spectroscopy detection.
出处
《光谱学与光谱分析》
SCIE
EI
CAS
CSCD
北大核心
2017年第8期2513-2518,共6页
Spectroscopy and Spectral Analysis
基金
国家自然科学基金项目(31271973)资助
关键词
近红外光谱
水分补偿
内部品质
无损检测
Near-infrared spectroscopy
Moisture compensation
Internal quality
Non-destructive detection
