基于近红外光谱分析技术的食品包装塑料的定性分析

Research on Food Packaging Plastics Based on Near Infrared Spectroscopy

该文基于近红外漫反射光谱分析技术对食品包装材料聚乙烯、聚丙烯进行定性判别试验研究,选取不同波段范围、采用不同光谱预处理方法,使用主成分分析法(Principal component analysis,PCA)结合SIMCA、贝叶斯判别、K-近邻3种模式识别方法建立定性预测模型,并根据正确识别率比较了各模型预测性能。结果表明:使用SIMCA方法、贝叶斯判别、K-近邻3种方法建立的定性校正模型均在1050~1550 nm波长范围内效果较好;采用矢量归一化、标准正态变量变换、中心化、滑动均值滤波、多项式平滑滤波、一阶微分6种光谱预处理方法和上述3种模式识别方法对塑料样品近红外光谱进行了数据处理,其中在1050~1550 nm范围内,主成分因子数为3,采用原始光谱建立的K-近邻定性校正模型较优,对样品校正集和预测集的正确识别率均为100%。可为食品包装材料聚乙烯、聚丙烯的快速鉴别研究提供参考。

Based on near infrared diffuse reflectance spectroscopy for the qualitative discrimination on food packaging materials polyethylene and polypropylene using different spectral pretreatment methods with different wavebands,principal component analysis(PCA)combined with three pattern recognition methods of SIMCA,Bayes discriminant and K-Nearest neighbor were adopted to establish three qualitative prediction models,and the prediction performances of the models were compared according to their correct recognition rates,in order to select the best model.Results showed that the qualitative correction models established by SIMCA,Bayes discriminant analysis and K-Nearest neighbor are better in the wavelength range of 1050-1550 nm.Six types of spectral preprocessing methods,i.e vector normalization,standard normal variable transformation,Centralization,moving average filtering,Savitzky-Golay filtering and first order differential combined with three pattern recognition methods of SIMCA,Bayes discrimination and K-Nearest neighbors were used to process the near-infrared spectra of plastic samples.In the range of 1050-1550 nm,the principal component factor was 3.The qualitative correction model by K-Nearest neighbor using the original spectrum was the best,whose correct recognition rates for the sample's calibration set and prediction set were both 100%.It could provide a reference for the rapid identification of polyethylene and polypropylene as food packaging materials.