基于可见近红外光谱的糖类别快速鉴别研究

Fast Discrimination of Varieties of Sugar Based on Spectroscopy Technology

提出了一种用近红外光谱技术快速鉴别糖类别的新方法。采用近红外光谱获取白砂糖、木糖醇、双歧糖和葡萄糖等四种糖类别的光谱反射特征曲线,采用偏最小二乘法进行模式特征分析,经过交互验证法判别,确定最佳主成分数为11。完成特征提取后,将11种主成分作为神经网络的输入变量,建立了3层BP神经网络。四个类别的糖样本数均为40,共计160个样本,将其分成训练集样本120个和预测集样本40个,对40个未知样本进行预测,准确率为100%。说明所提出的方法对于糖类别具有很好的分类和鉴别能力。

Visible and near-infrared reflectance spectroscopy （NIRS） was applied in the discrimination of sugar varieties. NIRS is a pollution-free, rapid, quantitative and qualitative analysis method, with the characteristics of high speed, non-destructiveness, high precision and reliable detection data, etc. Four kinds of sugar were gained from the local market and each species was divided into 40 samples. One hundred twenty samples were used as the training set and the remainders （total 40 samples） formed the prediction set. Samples were scanned by a spectroradiometer within a wavelength region of 325-1 075 nm. Three pre-processing methods were applied on the spectra prior to building the PLS regression model. The multivariable analysis using partial least square （PLS） was applied to abstract characteristics of the pattern. Through full cross validation, 11 principal components presenting important information of spectra were confirmed. The correlation coefficient （R）, residual variance （Rv） and standard error of calibration （SEC） were 0. 999 916, 0. 000 985 and 0. 014 538 respectively. Then, these 11 principal components were taken as the input of BP neural network. This model was used to predict the varieties of 40 unknown samples. Through training and prediction, the recognition rate of 100% was achieved by BP neural network. This model has come to be reliable and practicable. Thus, it is concluded that PLS analysis combined with BP neural network is an available alternative for pattern recognition based on the spectroscopy technology.