小波去噪对近红外光谱鉴别转基因菜籽油的影响分析

Effect of Wavelet De-noising to Identify Near Infrared Spectroscopy of Transgenic Rapeseed Oil

【目的】建立基于近红外光谱的转基因菜籽油定性鉴别模型,研究小波去噪对光谱的预处理对鉴别准确率的影响。【方法】利用前期已收集的鲁花、金龙鱼等6种品牌的瓶装或桶装的菜籽油共计117份样品,其中转基因菜籽油样品53份、非转基因菜籽油样品64份,采用德国BRUKER公司的MATRIX-F型傅里叶近红外光谱仪对这些样品进行全谱段的光谱采集;利用小波分析对菜籽油光谱数据进行预处理,选用db3小波对光谱进行软阈值去噪;在菜籽油样品近红外光谱数据的基础上,采用判别偏最小二乘法(DPLS)建立转基因菜籽油定性鉴别模型。【结果】对比小波去噪预处理前后转基因菜籽油鉴别的准确率,DPLS鉴别模型的准确率从96.43%提升到了100%。【结论】小波去噪预处理可以有效地提高近红外光谱转基因菜籽油鉴别模型的准确率。

[Objective] The aim of this study was to establish the transgenic rapeseed oil identification models based on near infrared spectra, and study the effect of spectral preprocessing with wavelet denoising on the identification accuracy rate. [Method] Luhua,Jinlongyu and other six brands of bottled or barreled rapeseed oil altogether 117 samples have been already collected in earlier stage,which include 53 samples of transgene rapeseed oil and 64 samples of non-transgene rapeseed oil. The full spectrum of spectrum data of the 117 samples was collected by the FT-NIR analyzer of BRUKER company of Germany; the rapeseed oil spectral data was preprocessed by wavelet analysis, using db3 wavelet soft threshold to denoise the spectra;based on the near infrared spectral data of rapeseed oil samples, discriminant partial least squares （DPLS） were applied to set up the identification model of modified rapeseed oil. [Results] This study compared the accuracy of the modeling methods before and after the wavelet preprocessing. The accurate rate of the DPLS model increased from 96.43% to 100%. [Conclusion] The results indicated that wavelet denoising pretreatment can improve the accuracy of near infrared spectra of transgenic rapeseed oil identification model effectively.