基于近红外光谱技术荞麦无损检测方法研究

发布时间：2018-04-18 18:30

本文选题：近红外光谱 + 荞麦　；参考：《昆明理工大学》2013年硕士论文

【摘要】：随着人民生活水平的日益提高,人们不再满足于食品的充足供应,越来越注重食物的功能性。不但在日常饮食过程中获得营养,又要达到以日常饮食养生、祛病的目的。荞麦应时而生,现阶段,荞麦品质优劣主要通过主观意识判别,缺乏科学高效的品质检测和分级技术,所以难以快速客观地鉴别荞麦品质和级别,导致各种荞麦制品的品质良莠不齐,难以打出优质荞麦品牌。常规荞麦品质检测方法快捷高效,但易受个人主观因素影响,很难标准化。因此在日常生活中,需要快速、简单、准确又易操作的荞麦品质检测方法来改变荞麦品质检测的落后情况。近红外光谱技术(Near Infared Spectroscopy, NIR)具有分析速度快、多组分同时测定、无需预处理、非破坏性分析、远距测定、实时分析,低分析成本和操作简单等优点。在农产品品质检测中正快速替代常规检测方法。采用近红外光谱技术采集荞麦光谱信息,研究荞麦内部化学特性与荞麦品质的相关关系,建立快速、准确、简便和稳定的荞麦品质评定方法,具有很好的理论研究意义和实际应用价值。本论文开展了基于近红外光谱技术荞麦无损检测方法研究。首先,使用近红外光谱技术结合最小二乘法分别进行荞麦蛋白质、淀粉和总黄酮含量预测,基于MATLAB平台下采用主成分分析和BP神经网络分别进行荞麦蛋白质、淀粉和总黄酮含量预测；然后,采用近红外光谱技术结合支持向量机算法对荞麦光谱信息进行品种鉴别；最后,基于中红外光谱技术结合主成分分析和BP神经网络分别进行荞麦蛋白质、淀粉和总黄酮含量预测。本论文的主要内容和结论如下： (1)采用FieldSpec3光谱仪采集荞麦光谱信息,使用平滑和多元散射校正两种方法对光谱信息进行预处理,采用近红外光谱技术分别建立了荞麦蛋白质、淀粉和总黄酮含量线性和非线性预测模型。首先采用近红外光谱技术结合最小二乘法分别建立荞麦蛋白质、淀粉和总黄酮含量预测模型,但预测集荞麦蛋白质、淀粉和总黄酮含量相关度均较低。然后,基于主成分分析进行主成分数提取,将提取主成分数与荞麦蛋白质、淀粉和总黄酮含量作为BP神经网络的输入和输出变量分别建立荞麦蛋白质、淀粉和总黄酮含量预测模型并进行预测,得出荞麦蛋白质、淀粉和总黄酮含量相关度为0.718、0.761和0.911,相对误差分别为7.82%、4.35%和8.26%。结论表明：采用近红外光谱技术结合主成分分析和BP神经网络分别建立荞麦蛋白质、淀粉和总黄酮含量预测模型,其中总黄酮相关度较高,能够达到荞麦总黄酮含量预测的目的,采用最小二乘法进行荞麦蛋白质、淀粉和总黄酮含量预测是不可行的。 (2)采用近红外光谱技术结合支持向量机算法对荞麦光谱信息进行品种鉴别中,在MATLAB平台下采用LIBSVM软件包对荞麦光谱进行品种鉴别,训练了7个支持向量机,对8类不同产地荞麦预测率平均达到92.5%。结论表明：采用近红外光谱技术结合支持向量机算法对荞麦光谱进行品种鉴别是可行的。 (3)采用中红外光谱技术结合主成分分析和BP神经网络分别建立荞麦蛋白质、淀粉和总黄酮含量预测模型,结果表明：荞麦蛋白质、淀粉和总黄酮含量预测相关度分别是0.769、0.848和0.938,相对误差分别是9.31%、4.53%和8.53%,其中淀粉和总黄酮预测率较高,能够达到荞麦淀粉和总黄酮含量预测的目的。相对于近红外光谱技术,中红外光谱技术预测精度高于近红外光谱技术预测精度,但中红外光谱技术属于有损检测,不利于实现农产品快速检测。综上所述,本论文系统讨论了基于近红外光谱技术结合化学计量法进行荞麦组分预测和品种鉴别。在荞麦组分预测中,采用近红外光谱技术结合最小二乘法和BP神经网络分别对荞麦蛋白质、淀粉和总黄酮含量进行预测。在荞麦品种鉴别中,采用近红外光谱技术结合支持向量机算法对荞麦光谱信息进行品种鉴别,采用中红外光谱技术结合主成分分析和BP神经网络分别对荞麦蛋白质、淀粉和总黄酮含量进行预测,旨在完善荞麦近红外光谱检测方法,本项研究成果能为拥有自主知识产权荞麦及其他产品内在品质检测方法提供了基础理论和方法借鉴,具有很好的理论研究意义和实际应用价值。
[Abstract]:In order to improve the quality and quality of buckwheat , it is difficult to quickly and objectively identify the quality and grade of buckwheat , and it is difficult to standardize the quality and quality of buckwheat .
Then , using near - infrared spectroscopy combined with support vector machine algorithm to identify buckwheat spectral information ;
Finally , the content of buckwheat protein , starch and total flavonoids was predicted based on mid - infrared spectroscopy combined with principal component analysis and BP neural network . The main contents and conclusions of this thesis are as follows :

The content of buckwheat protein , starch and total flavone in buckwheat protein , starch and total flavonoids was predicted by near - infrared spectroscopy combined with the principal component analysis and BP neural network . The results showed that the correlation between the total flavonoids and total flavonoids was 0.718 , 0.761 and 0.911 , respectively . The relative error of buckwheat protein , starch and total flavonoids was 7.82 % , 4.35 % and 8.26 % respectively .

( 2 ) Using the near - infrared spectroscopy combined with the support vector machine algorithm to identify the buckwheat spectral information , using the LIBSVM software package to identify the buckwheat spectra under the MATLAB platform , seven support vector machines were trained , and the prediction rate of buckwheat in different regions of the eight categories reached 92.5 % . The conclusion shows that the identification of the buckwheat spectra by the near infrared spectroscopy combined with the support vector machine is feasible .

( 3 ) The prediction model of buckwheat protein , starch and total flavonoids was established by mid - infrared spectroscopy combined with principal component analysis and BP neural network . The results showed that the predictive correlation of buckwheat protein , starch and total flavonoids was 0 . 769 , 0 . 848 and 0 . 938 , respectively . The relative error was 9.31 % , 4.53 % and 8.53 % , respectively .

In conclusion , the content of buckwheat protein , starch and total flavone is predicted by near - infrared spectroscopy combined with least square method and BP neural network . In the identification of buckwheat variety , the content of buckwheat protein , starch and total flavone is predicted by using near infrared spectroscopy combined with support vector machine algorithm .

【学位授予单位】：昆明理工大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：S517

【参考文献】