Di Wu, Feng Lei, Zhang Chuan-Qing, He Yong
College of Biosystems Engineering and Food Science, Zhejiang University, Hangzhou 310029, China.
Guang Pu Xue Yu Guang Pu Fen Xi. 2007 Nov;27(11):2208-11.
Visible and near-infrared reflectance spectroscopy (Vis/NIRS) technique was applied to the detection of disease level of grey mold on tomato leave. Chemometrics was used to build the relationship between the reflectance spectra and disease level. In order to decrease the amount of calculation and improve the accuracy of the model, principal component analysis (PCA) was executed to reduce numerous wavebands into several principal components (PCs) as input variables of BP neural network. The loading value of PC1 was applied to qualitatively analyze which wavebands were more important for disease detection. Prediction results showed that when the number of primary PCs was 8 and the hidden nodes of BP neural network were 11, the detection performance of the model was good as correlation coefficient (r) was 0.930 while standard error of prediction (SEP) was 0.068 7. Thus, it is concluded that spectroscopy technology is an available technique for the detection of disease level of grey mold on tomato leave based on chemometrics used for data analysis.
可见/近红外反射光谱（Vis/NIRS）技术被应用于番茄叶片灰霉病病情程度的检测。采用化学计量学方法建立反射光谱与病情程度之间的关系。为了减少计算量并提高模型的准确性，执行主成分分析（PCA）将众多波段缩减为几个主成分（PC）作为BP神经网络的输入变量。利用PC1的载荷值定性分析哪些波段对病害检测更为重要。预测结果表明，当主成分数量为8且BP神经网络的隐藏节点数为11时，模型的检测性能良好，相关系数（r）为0.930，预测标准误差（SEP）为0.068 7。因此，得出结论：基于用于数据分析的化学计量学，光谱技术是检测番茄叶片灰霉病病情程度的一种可行技术。
