作　　者： ; ; ; ; (吴从元）;

机构地区： 浙江农林大学农业与食品科学学院

出　　处： 《传感技术学报》 2011年第6期928-933,共6页

摘　　要： 选取人工陈化山核桃(0 d、2 d4、d和6 d)和自然陈化山核桃(0 y、1 y和2 y)作为研究对象,采用德国PEN2便携式电子鼻进行检测,在主成分分析(PCA)的基础上,采用概率神经网络(PNN)模式识别方法建立山核桃陈化时间鉴别模型,模型参数Spread和主成分数通过交互验证的方法优化。结果标明,PCA基本可区分不同陈化时间的人工陈化和自然陈化的山核桃。当主成分数为4和Spread=0.1或Spread=0.2时,人工陈化山核桃所得识别模型最佳,校正集样本识别率为100%,预测集样本识别率为65%;当主成分数为2和Spread=0.1～0.6时,自然陈化山核桃所得识别模型最佳,校正集样本和预测集样本识别率均为100%。研究表明,基于主成分分析(PCA)和概率神经网络(PNN)的电子鼻技术可较好鉴别不同陈化时间的人工陈化和自然陈化的山核桃,且对自然陈化山核桃的识别效果要优于人工陈化山核桃。 Walnut（carya cathayensis Sarg）samples with artificial aging times（0 d,2 d,4 d and 6 d）and natural aging times（0 year,1 year and 2 year）were measured by PEN2 electronic nose made in Germany.Based on principal component analysis（PCA）,Probabilistic Neural Network（PNN）algorithm pattern recognition method was applied to build discriminating model from aging times.The parameter of Spread in the PNN model and the number of principal component factors were optimized by cross-validation method.The results showed that PCA could basically distinguish walnut of artificial aging times and natural aging times.Experimental results showed that the optimal recognition model of artificial aging walnut was obtained with 4 principal component factors and Spread=0.1 or 0.2,the discriminating rates of walnut in the calibration sets and prediction sets were 100% and 65%,respectively.However,the optimal model of natural aging walnut was obtained with 2 principal component factors and Spread=0.1~0.6 and both discriminating rates in the calibration and prediction sets were 100%.The present study demonstrated that Electronic nose technique with PCA and PNN methods could discriminate walnut well from diverse of artificial or natural aging times,and these methods were much better for natural aging walnut than those of artificial aging walnut.

关 键 词： 电子鼻 陈化时间 主成分分析 概率神经网络 山核桃

领　　域： [自动化与计算机技术] [自动化与计算机技术]