苹果质地品质近红外无损检测和指纹分析

Nondestructive measurement and fingerprint analysis of apple texture quality based on NIR spectra

为了探索近红外光谱快速无损检测苹果质地品质的方法,采集240个苹果样本的近红外光谱(波长800～2500nm),通过解析光谱图和进行不同的预处理,利用偏最小二乘法(PLS)和多元线性回归(MLR)建立回归模型和确定特征指纹图谱。基于波长范围为1300～2500nm,PLS结合多元散射校正(MSC)所建模型的预测效果最好,硬度模型的预测标准偏差(RMSEP)和决定系数(R2)分别为0.226kg/cm2、96.52%,脆度模型的RMSEP和R2分别为0.243kg/cm2、97.15%。用权重法基于PLS模型选择的硬度特征波长为1657、1725、1790、2455、1929、2304nm,脆度特征波长为1613、1725、1895、2304、2058、2087、2396nm,经MLR模型检验,特征波长与苹果的硬度和脆度有很高的相关性,硬度的RMSEP和R2分别为0.271kg/cm2、90.30%,脆度的RMSEP和R2分别为0.304kg/cm2、91.64%。结果表明,PLS模型和特征指纹光谱均能准确预测苹果的质地品质,为苹果质地品质的评价提供了快速、直观、简便、可行的新方法。

A rapid and nondestructive way to measure texture of apple was put forward based on NIR spectra and the relationships between NIR spectra and firmness and crunchiness were developed. The NIR spectra were acquired from 240 samples of apples with the wavelength from 800 to 2500nm. The multivariable analyses including partial least squares （PLS） and multiple linear regressions （MLR） were conducted to build the regression models and select the fingerprint spectra of firmness and crunchiness. The excellent models with high coefficient of determination （R2： 96.52%; 97.15 % ） and low RMSEP （0.226 kg/cm^2; 0.243 kg/cm^2） were obtained by PLS＋MSC models based on wavelength from 1300 to 2500 nm. The loading weights from PLS model were found to be the sensitive firmness wavelengths （1657, 1725, 1790, 2455, 1929 and 2304 nm） and crunchiness wavelengths （1613, 1725, 1895, 2304, 2058, 2087 and 2396 nm）. These wavelengths were strongly related with apple＇s texture（r： 0.921, 0.957） by MLR models evaluated. The results indicate that the PLS models and the fingerprint spectra can predict apple texture quality accurately, A new method which can evaluate apple texture quality rapidly, visually, simply and feasibly was developed.