Yu Jia-jia, He Yong, Bao Yi-dan
College of Biosystems Engineering and Food Science, Zhejiang University, Hangzhou 310029, China.
Guang Pu Xue Yu Guang Pu Fen Xi. 2008 Dec;28(12):2839-42.
Mango is a kind of popular tropic fruit in the word, and its quality will affect the health of consumers. Unsaturated acid is an important component in mango. So it is very important and necessary to detect the sugar content and valid acidity in mango fast and non-destructively. Visible and short-wave near-infrared reflectance spectroscopy (VIS/SWNIRS) was applied in the present study to predict sugar content and valid acidity of mango. Because of the non-linear information in spectral data characteristics of the pattern were analyzed by neural network optimized by genetic algorithm (GA-BP). Spectral data were compressed by the partial least squares (PLS). The best number of principal components (PCs) was selected according the accumulative reliabilities (AR). PCs could be used to replace the complex spectral data. After some preprocessing and through full cross validation, 17 principal components presenting important information of spectra were confirmed as the best number of principal components for valid acidity, and 18 PCs as best number of principal components for sugar content. Then, these best principal components were taken as the input of GA-BP neural network. One hundred thirty five samples were randomly collected as modeling, and the remaining 45 as samples to check the forecast results by the model. For the sake of testing the GA-BP model, at the same time we took the BP neural network on the same PCs. The quality of the calibration model was evaluated by the correlation coefficients (R) and standard error of calibration (SECV), and the prediction results were assessed by correlation coefficients (R) and standard error of prediction (SEP). Comparing PLS-BP model with PLS-GA-BP model, the coefficients of determination (R) of 0.788/0.83699 and standard errors of prediction (SEP) of 0.133312/0.109447 were calculated in valid acidity. The sugar content result was calculated by the coefficients of determination (R) = 0.75705/0.85409 and standard errors of prediction (SEP)0.864676/0.60934. Thus, it is obvious that this model is reliable and practicable. And the PLS-GA-BP model based on the spectroscopy technology is a better pattern to predict sugar content and valid acidity of mango, giving a new method for detecting fruit's sugar content and valid acidity.
芒果是世界上一种受欢迎的热带水果，其品质会影响消费者的健康。不饱和酸是芒果中的重要成分。因此，快速、无损地检测芒果中的糖分含量和有效酸度非常重要且必要。本研究应用可见/短波近红外反射光谱法（VIS/SWNIRS）预测芒果的糖分含量和有效酸度。由于光谱数据具有非线性信息，采用遗传算法优化的神经网络（GA-BP）分析模式特征。通过偏最小二乘法（PLS）对光谱数据进行压缩。根据累积可靠性（AR）选择最佳主成分数（PCs）。PCs可用于替代复杂的光谱数据。经过一些预处理并通过全交叉验证，确定17个呈现光谱重要信息的主成分作为有效酸度的最佳主成分数，18个PCs作为糖分含量的最佳主成分数。然后，将这些最佳主成分作为GA-BP神经网络的输入。随机收集135个样本进行建模，其余45个作为样本通过该模型检验预测结果。为了测试GA-BP模型，同时在相同的PCs上采用BP神经网络。通过相关系数（R）和校准标准误差（SECV）评估校准模型的质量，通过相关系数（R）和预测标准误差（SEP）评估预测结果。将PLS-BP模型与PLS-GA-BP模型进行比较，有效酸度的决定系数（R）分别为0.788/0.83699，预测标准误差（SEP）分别为0.133312/0.109447。糖分含量结果的决定系数（R） = 0.75705/0.85409，预测标准误差（SEP）为0.864676/0.60934。因此，显然该模型可靠且实用。基于光谱技术的PLS-GA-BP模型是预测芒果糖分含量和有效酸度的较好模式，为检测水果的糖分含量和有效酸度提供了一种新方法。
