应用近红外光谱技术快速检别酱油品牌的研究

提出了一种采用近红外光谱快速鉴别酱油品牌的新方法,对不同品牌的酱油建立相应的指纹模型。对市场上8种典型品牌的酱油,通过近红外透射获取光谱曲线,选择了其中噪声较小的7625～3684cm-1共3942个波段作为建模分析数据。为了减少原始数据量,提高数据处理效率,对原始数据进行了多项式平滑拟合等预处理,采取主成分分析法,得到能反映酱油99·99%光谱信息的8个主成分。由这8个主成分得到的得分图,可以区分其中某几个品牌,但是不能做到区分全部品种,因此选取了人工神经网络进行了进一步信息提取与种类判别。将8个主成分作为人工神经网络的输入,对应的酱油品牌作为输出,通过不断调整参数,建立了最优的BP神经网络。8个品牌共242个样本作为建模学习样本,每个品牌各10个共80个样本作为检验样本。结果表明,在0·98的置信区间里取得了98·75%的识别正确率,为不同等级和品牌的酱油鉴别提供了一种新的方法。

Study on Fast Discrimination of Brands of Soy Sauce Using Near Infrared Spectra

A new method for the fast discrimination of brands of soy sauce by means of near infrared spectroscopy(NIRS) was developed and these eight kinds of soy sauce had got its "identity card".The experiment adopted typical eight brands of soy sauce which we bought in the market.Total 3 942 frequencies from 7 625 to 3 684 cm-1 transmit wavelength were gotten to set up a analyses model.In order to handle these data efficiently,after pretreatment,firstly,principal component analysis(PCA) was used to compress thousands of spectral data into several variables and to describe the body of the spectra,the analysis suggested that the cumulate reliabilities of the first eight components was more than 99.99%.According to the first eight components,the authors could distinguish some brand of the soy sauce but could not deal with all of them.So the authors chose ANN-BP as further research method.The eight components were secondly applied as ANN-BP inputs.The experiment took total 242 examples of 8 kinds of soy sauce as original model examples and left 10 every kind as unknown samples to predict.Finally,the result indicated the distinguishing rate is 98.75% in 0.98 reliable area.This paper could offer a new method to the discrimination of varieties of soy sauce.