葡萄浆果糖度可见/近红外光谱检测的研究

针对可见/近红外光谱与水果糖度存在非线性相关的特点，利用漫反射光谱测定方法获取了葡萄浆果的可见/近红外光谱，提出了应用偏最小二乘（PLS）结合人工神经网络（ANN）建立葡萄浆果糖度的预测模型，利用偏最小二乘法（PLS）对原始光谱数据进行处理，得出交叉检验的最佳主因子数为3，并将3个主因子的得分作为三层BP神经网络的输入。通过定标集样本对BP神经网络进行训练，用优化的BP神经网络模型对预测集样本进行预测。PLS-ANN模型对样本的预测模型检验参数r2为0.908，RMSEP为0.112，Bias为0.013，好于只使用PLS模型的预测模型检验参数r2为0.863，RMSEP为0.171, Bias为0.024。结果表明，利用近红外光谱技术无损检测葡萄浆果糖度等内部品质是可行的，为今后进一步分析建立浆果内部品质预测模型奠定了基础。

Research on the Sugar Content Measurement of Grape and Berries by Using Vis/NIR Spectroscopy Technique

Aiming at the nonlinear correlation characteristic of Vis/NIR spectra and the corresponding sugar content of grape and berries, the Vis/NIR spectra of grape and berries were obtained by diffusion reflectance. A mixed algorithm was presented to predict sugar content of grape and berries. The original spectral data were processed using partial least squares (PLS), and three best principal factors were selected based on the reliabilities. The scores of these 3 principal factors would be taken as the input of the three-layer back-propagation artificial neural network (BP-ANN). Trained with the samples in calibration collection, the BP-ANN predicted the samples in prediction collection. The values of decision coefficient (r2), the root mean squared error of prediction (RMSEP), and bias were used to estimate the mixed model. The observed results using PLS-ANN (r2=0.908, RMSEP=0.112 and Bias=0.013) were better than those obtained by PLS (r2=0.863, RMSEP=0.171, Bias=0.024). The result indicted that the detection of internal quality of grape and berries such as sugar content by nondestructive determination method was very feasible and laid a solid foundation for setting up the sugar content forecasting model for grape and berries.