摘要
[目的]采用近红外光谱技术法,快速鉴别茶油掺伪。[方法]基于近红外光谱技术,比较马氏距离聚类分析法与反向传播神经网络,建立茶油与掺有菜籽油、棕榈油掺伪茶油的模式识别模型。[结果]采用马氏距离聚类分析法建模时,参数如下:光谱一阶导数处理后,结合SNV、Norris Derivative滤波方法,经主成分分析法,提取8个主成分,模型对预测集样本的准确率达100%;采用反向传播神经网络建模时,参数如下:输入向量为前8个主成分的33个吸收峰,隐含层神经元个数为15,训练学习速度为0.1,训练220步时,模型对预测集样品识别准确率亦为100%。[结论]反向传播神经网络方法更加具有较快的运算速度和较好的收敛性,可为茶油品质评价与检测提供一种新方法。
[Objective]Using near-infrared spectroscopy technology to quickly identify camellia oil adulteration.[Method]Based on near infrared spectroscopy,two pattern recognition models were developed for discriminating camellia oil and adulterated oil with rapeseed oil and palm oil,which were bulit by Mahalanobis distance discriminative model and Back-Propagation(BP)network.[Result]When modeling with Mahalanos distance clustering,the parameters were as follows:first derivative spectrum combined with standard normal variate(SNV)and Norris Derivative,8 principal components compressed from the original data processed by PCA,the model’s accuracy of discrimination in the prediction set was 100%.When modeling with BP neural networks,the parameters were as follows:33 absorption peaks data,8 principal components processed by PCA were taken as inputs of the BP Network,the number of hidden neurons was 15,learning rate was 0.1,training steps were 220,the BP model was built for identification of camellia oil and adulterated oil,and the model’s recognition correct rate was 100%.[Conclusion]The BP network has rapider operation speed,better convergence,which provides a new method for the quality evaluation and determination of camellia oil.
作者
荣菡
罗懿
黄镘淳
RONG Han;LUO Yi;HUANG Man-chun(Zhuhai School of Material and Environment,Beijing Institute of Technology,Zhuhai,Guangdong 519088)
出处
《安徽农业科学》
CAS
2019年第19期204-206,219,共4页
Journal of Anhui Agricultural Sciences
基金
广东高校省级重点平台及重大科研项目(青年创新人才类)(2017KQNCX250)
广东省大学生创新创业项目(201813675027)
关键词
近红外光谱
模式识别
马氏距离
反向传播神经网络
茶油
掺伪油
Near infrared spectroscopy
Pattern recognition
Mahalanobis distance
Back-propagation network
Camellia oil
Adulterated oil
