基于可见-近红外光谱的鲜食葡萄成熟品质关键指标检测

酚类物质是评价葡萄成熟品质的重要指标，本文利用可见-近红外光谱技术结合化学计量学定量分析方法对葡萄皮总酚、籽总酚、皮单宁和籽单宁含量开展了无损检测研究。通过手持式可见-近红外光谱仪采集巨玫瑰葡萄波长400～1 029 nm范围内的漫反射光谱，采用SPXY算法将其划分为校正集和预测集，结合标准正态变换(Standard normal variate, SNV)、多元散射校正(Multiplicative scatter correction, MSC)、一阶导数(First derivative, 1D)、二阶导数(Second derivative, 2D)、Savitzky-Golay卷积平滑(Savitzky-Golay smoothing,SG)和Savitzky-Golay卷积平滑+一阶导数(SG+1D)6种预处理方法以及偏最小二乘回归(Partial least squares regression, PLSR)、支持向量机回归(Support vector machine regression, SVR)和卷积神经网络(Convolutional neural networ...

Detection of Key Indicators of Ripening Quality in Table Grapes Based on Visible-near-infrared Spectroscopy

Phenolic compounds play a crucial role in assessing the internal quality of grapes and hold significant importance in this regard. The capability of visible-near-infrared (Vis-NIR) spectroscopy combined with multivariate regression models was explored to detect the contents of total phenolics and tannins in grape skins and seeds. Reflectance spectra data of Muscat Kyoho grapes were collected within the wavelength range of 400nm to 1029nm, and the samples were divided into correction set and prediction set by SPXY algorithm. Six commonly used preprocessing methods such as standard normal variate (SNV), multiplicative scatter correction (MSC), first derivative (1D), second derivative (2D), Savitzky-Golay smoothing (SG) and SG+1D were applied to the spectral data, and the competitive adaptive reweighted sampling algorithm (CARS) was utilized to select informative wavelengths. The quantitative models for comprehensive analysis of total phenolics and tannins in grape skins and seeds based on full spectra and effective wavelengths were established by partial least squares regression (PLSR), support vector machine regression (SVR), and convolutional neural network (CNN). The results showed that for the total phenolics in grape skins, total phenolics and tannins in grape seeds, the models on the basis of effective wavelengths performed better than those with full spectra data. While for the tannins in grape skins, the models constructed with full spectra yielded better results than the feature wavelength-selected models. The optimal models for the total phenolics and tannins in grape skins and seeds were RAW-CARS-SVR, 1D-CARS-SVR, RAW-CNN and RAW-CARS-PLSR, respectively. The correlation coefficent of calibration set (Rc) were 0.96, 0.99, 0.96 and 0.91, the correlation coefficent of prediction set (Rp) were 0.95, 0.99, 0.83 and 0.89, the residual predictive deviation (RPD) were 3.56, 7.30, 1.92 and 2.25, respectively. Therefore, the developed method could realize the non-destructive detection of the contents of total phenolics and tannins in grape skins and seeds.