基于电子鼻和神经网络的牛肉新鲜度的检测

Application of Electronic Nose and Neural Network in Beef Freshness Detection

本文对一批新鲜牛肉进行感官评价、挥发性盐基氮检测、微生物含量检测和电子鼻检测，检测时间为冷藏第0、3、5、7、10、12、14d。应用电子鼻第50s的响应信号建立牛肉样品的模式识别分析模型。运用马氏距离分析牛肉样品新鲜度的变化，样品与新鲜样品间的马氏距离随冷藏时间的延长而增大，在冷藏5d后，样品的质量有明显跃变；运用主成分分析和线性判别分析区分不同冷藏时间的样品，除第肚5d样品有部分重合外，其他天数样品都能很好区分；利用逐步判别分析进行冷藏时间判别，正确率为96．19％；分别采用BP神经网络（BPNN）和广义回归网络（GRNN）建立牛肉冷藏天数与感官理化指标间的相关模型，实验表明GRNN模型效果较好，该模型对冷藏时间、挥发性盐基氮（TVBN）、细菌总数和感官得分的预测误差Se分别为1．36d、4．64×10^-2mg/g、1．61×10^-6cfu／g和1.31。

Fresh beef under 0~ 12 day cold storage were detected by sensory evaluation, semi-micro determination of nitrogen detection, microbiological testing and electronic nose （e-nose） detection. For the e-nose detection, the 50th s sensor signals were extracted for analysis. Mahalanobis Distance （MD） between fresh and stored samples enlarged with the increase of storage time, and the quality of beef significantly changed after stored for 5 days. Principle component analysis （PCA） and Linear Discriminant Analysis （LDA） results indicated that all the beef samples could be well distinguished except a little overlap between the samples stored for 0~5 days. Stepwise Linear Discriminant Analysis （Step-LDA） was applied to predict the storage time and got an accuracy of 96.19%. Back Propagation Neural Network （BPNN） and Generalized Regression Neural Network （GRNN） were also used to build a correlation model between the storage time and the physicochemical indicators, which showed that GRNN was better than BPNN. The prediction error based on the GRNN model for the storage time, total volatile basic nitrogen, microbial population and sensory scores were 1.36 days, 4.64×10^-2 mg/g, 1.61×10^-6 cfu/g and 1.31, respectively.