摘要
采用PEN2型电子鼻系统对芝麻油的玉米油掺假进行定性鉴别和定量预测,运用主成分分析,逐步判别分析和Fish-er线性判别函数变换对原始数据进行预处理,从而降低原始数据空间的维数,并用判别分析与人工神经网络对数据进行进一步分析,考察了不同的数据预处理方法的效果。判别分析结果表明,采用Fisher线性判别函数变换所得到的十个变量判别能力最强,误判率为0.61%,仅有1个样品出现误判。在BP神经网络的定量预测中,采用逐步判别分析所筛选出的十个变量作为网络输入,所得的预测结果最为理想,绝对误差个体值的95%置信区间最小,为(-4.71%,3.38%),均方误差为4.75,预测值与实际值之间有极显著的相关性,相关系数R=0.99808。
An electronic nose was used to detect the adulteration of sesame oil with corn oil. In order to reduce the dimension of the data matrix, three different variable selection techniques were employed： PCA, step-LDA and Fisher linear transformation. And then the pattern recognition technique of linear discriminant analysis （LDA） and artificial neural network （ANN） were used to check the effect of the three dimension reduction methods mentioned above. In the process of LDA , the Fisher linear transformation is most effective, the error rate was 0.61%, only 1 sample was misclassified; in the process of ANN, the results obtained using the ten variables selected by step-LDA were more acceptable than others, the 95% individual confidence interval is （-4. 71%,3. 38%）, the mean square of error was 4. 75, and the correlation between predicted concentration and genuine concentration is 0. 998 08.
出处
《传感技术学报》
EI
CAS
CSCD
北大核心
2006年第3期606-610,共5页
Chinese Journal of Sensors and Actuators
基金
国家教育部新世纪人才支持计划资助(NCET-04-0544)
国家自然科学基金资助(30571706)
关键词
电子鼻
主成分分析
逐步判别分析
BP神经网络
electronic nose
principal component analysis
stepwise linear discriminant analysis
BP-neuralnetwork.
