主成分-人工神经网络在近红外光谱定量分析中的应用

Application of Principal Components-Artificial Neural Networks in Near _ infrared Spectroscopy Quantitative Analysis

近红外光谱的主成分由非线性迭代偏最小二乘法（ＮＩＰＡＬＳ）求出。主成分作标准化处理后，作为Ｂ－Ｐ神经网络的输入结点进行非线性迭代。该法的优点是，充分利用了全光谱的数据，得到消除噪声后的最佳主成分，能建立非线性模型，Ｂ－Ｐ神经网络迭代时间显著缩短。用该法对大麦中的淀粉含量进行了定量分析研究。结果为：校准和预测的相关系数分别为０．９８１和０．９５３，校准和预测的相对标准偏差分别为１．７０％和２．４８％。

The principal components of near _ infrared spectroscopy were calculated by nonlinear iterative partial least squares(NIPALS).After standardization,these principal components were used as input nodes of back propagation artificial neural networks(B-P ANN). ANN was used to build nonlinear model. The data of whole spectra can be fully utilized, the best principal components free from noise and nonlinear model obtained, and the iterative time of B-P ANN shorted strongly in the method. The method has been applied to quntitatively determine the starch of barley. The calibration and prediction correlation coefficients are 0.981 and 0.953, the relative standard deviations are 1.70% and 2.48%, respectively.