小波变换-可见-近红外光谱技术用于鉴别品牌料酒的研究

Study on discrimination of brands of seasoning wine using Vis-near infrared spectra

模式识别技术广泛应用于食品种类、品牌和原产地的分类鉴别。本文测定了三个品牌114个料酒样品的可见-近红外光谱,利用小波变换技术对光谱信号进行了去噪和压缩处理,并采用Fisher权重法计算了16个小波细节系数的Fisher权重。以16个小波细节系数为特征变量采用向量相似度法对三种不同品牌料酒进行了相似度分析,主成分分析法能显著区分三种不同品牌料酒。偏最小二乘(PLS)、径向基人工神经网络(RBF-ANN)和Fisher线性判别(LDA)三个判别模型对预报组的料酒品牌进行鉴别,三个判别模型对料酒品牌鉴别的准确率均达到100%。

Discrimination of kinds,brands and origins of food products is a very active area with the application of pattern recognition techniques.The Vis-near infrared spectroscopic signals of 114 seasoning wine were measured and compressed by wavelet transform,and the Fisher ratios of 16 wavelet detail coefficients were calculated by Fisher F-ratio approach.The similarities of three different brands of seasoning wine samples were calculated by vector similarity analysis using 16 wavelet detail coefficients.Principal component analysis showed significant difference among seasoning wine of three brands.The partial least square（PLS）,radial basis function-artificial neural network（RBF-ANN）and linear discriminant analysis methods（LDA）were used to develop discriminant model and 100%accuracy of prediction was obtained for the samples in the verification set.