基于电子鼻和电子舌的羊肉品质检测

【作者】 田晓静；

【导师】 王俊；

【作者基本信息】 浙江大学 ， 生物系统工程， 2014， 博士

【摘要】 我国已成为羊肉生产和消费大国,于2010年羊肉产量已达到410万吨,占世界羊肉产量的1/3,同时羊肉的消费量年均增长10%。由于产地和消费市场之间的地理差异,羊肉多以冻藏贮藏、运输和分销以调节肉品市场。但冷链条件的不完善造成冻藏肉受到反复解冻和冻结的影响,使其食用品质和经济价值降低。此外,以低价值肉掺入或替代羊肉的现象时有发生,对市场监管问题及消费者的身体健康带来了严重影响,并严重干扰广大穆斯林群众肉类消费和宗教信仰。本文围绕电子鼻和电子舌技术在羊肉品质检测中的应用展开研究。采用电子鼻和电子舌系统检测和识别了羊肉、猪肉、鸡肉、冻融不同次数的羊肉及混入猪肉、鸡肉的掺假羊肉糜样品,实现了电子鼻、电子舌及电子鼻和电子舌联用信号对不同种类样品的区分和识别；建立了羊肉冻融次数和羊肉糜中混入的猪肉、鸡肉含量的有效预测模型。主要结论如下：(1)通过单因素方差分析和主成分分析对样品的分类效果,获得电子鼻的较佳检测条件为：样品量10g、顶空体积250mL、载气流速200mL/min和顶空生成时间30min。电子舌检测羊肉浸提液的较佳试验条件为：以100mL0.1mol/L氯化钾溶液浸提15g羊肉糜样品30min获得的羊肉浸提液进行检测。在上述优化的检测条件下,以羊肉、猪肉和鸡肉为样品进行电子鼻和电子舌检测,主成分分析和判别分析结果表明电子鼻、电子舌均具有区分不同种类生肉(猪肉、鸡肉和羊肉)的能力。(2)采用电子鼻和电子舌对冻融不同次数的羊肉进行了监测,同时检测了其理化指标(挥发性盐基氮、pH、色度L*值、a*值和b*值)的变化情况。不管是电子鼻、电子舌或两者直接联用的信号,主成分分析和典则判别分析都成功的区分了不同冻融次数的羊肉；以电子鼻、电子舌或两者的联用信号为输入,采用偏最小二乘回归分析、主成分回归分析和多元线性回归分析对羊肉冻融次数和理化指标进行预测,三种回归方法建立的定量预测模型均成功预测了羊肉的冻融次数和理化指标(除b*)。电子鼻、电子舌及其联用信号均能实现对冻融不同次数羊肉的快速鉴别,且能成功预测冻融次数和理化指标：且采用联用信号的判别和预测效果均优于单独采用电子鼻、电子舌单独进行分析结果。(3)应用电子鼻、电子舌快速鉴别混入不同比例猪肉、鸡肉的掺假羊肉糜及其浸提溶液。对电子鼻检测信号进行分析时,以判别分析结果的正确率为依据,优化传感器阵列组合方式和特征参数提取(逐步判别分析分析、主成分分析、loading分析)。以优化后参数进行分析时,主成分分析和典则判别分析可成功区分(正确率高达99.17%)不同掺假羊肉样品；采用多元线性回归分析、偏最小二乘回归分析、BP神经网络及主成分回归分析建立掺假羊肉中混入其他成分的含量的定量预测模型时,回归模型的R2>0.9,预测误差均在10%以内。对掺假羊肉的电子舌信号进行分析时,主成分分析可以基本能实现对混入不同含量猪肉、鸡肉的掺假羊肉实现区分,典则判别分析的判别结果比主成分分析结果略好。建立的混入猪肉、鸡肉的含量定量预测模型的相关系数R2高达0.99,预测值均方根误差均小于3%,成功预测了掺假羊肉中掺杂物的含量。采用电子舌定性判别不同掺假羊肉和定量预测掺假羊肉中混杂其他成分的含量时,其检测效果优于电子鼻。(4)对掺假羊肉的电子鼻和电子舌响应信号,研究了直接联用、逐步判别分析优选参数、主成分特征提取等方法融合电子鼻和电子舌信号,并对比电子鼻、电子舌及其联用信号的判别和预测效果。发现任一融合了气味和滋味信息的电子鼻和电子舌联用信号方法均能提高掺假羊肉的判别正确率和预测的精度,在此基础上建立了基于联用信号的快速鉴别羊肉掺假和定量预测掺假物含量模型,成功预测了掺假羊肉糜中混入猪肉、鸡肉的含量。相比于单一气味或滋味信息的分析,电子鼻和电子舌数据的联用融合了气味和滋味信息,使综合分析的结果更全面,有利于提高区分鉴别的正确率和定量预测模型的有效性。更多还原

【Abstract】 The production of mutton in China has reached4.1million tons in2010, occupying a third of the world production of mutton. China has been a great nation of mutton production and consumption, with a growing consumption by an average growth rate of10%a year. Mutton is mostly frozen stored, transported and distributed to adjust meat market due to the geographical differences between producer and consumer market. The eating quality and economic value of mutton is greatly affected by the repeated thawing and freezing cycles caused by the imperfect conditions of cold chains in the processing, transportation and distribution process. What’s more, Adulteration of meat, involving the replacement of selected breeds, particular geographical region or particular traditional method with other cheaper animal proteins and even none meat proteins (soy proteins), has jeopardized the market regulation and consumers’ health, and religion belief of Muslims. Electronic nose and electronic tongue were employed to detect samples of mutton, pork, chicken, mutton with different freeze-thaw cycles, and minced mutton adulterated with different content of pork or chicken. Combined with pattern recognition methods, meat samples were classified according to their origin, and the physicochemical parameters and content of adulterants in minced meat were predicted using predictive models. The main conclusions are as follows:With the help of One-way analysis of variance (ANOVA) and principle component analysis (PCA), the optimum experimental parameters were acquired with15g of mutton sample extracted by100mL K.C1solution for electronic tongue,10g sample with30min headspace-generated time in250mL beaker with a flow rate of200mL/min for electronic nose. With the optimized experimental parameters, meat samples of pork, chicken and mutton were detected and analyzed by PCA and canonical discriminant analysis (CDA). Results were obtained that meat of different kinds could be discriminated both by electronic tongue and electronic nose.Electronic nose, electronic tongue and physicochemical parameters detection of total volatile basic nitrogen (TVB-N), pH, Color values of L*, a*and b*were employed to monitoring the effect of freeze-thaw cycles on mutton. Mutton samples with different freeze-thaw cycles were successfully differentiated by electronic nose, electronic tongue and their fusion data by PCA and CDA. Cycles of mutton freeze-thawed and the physicochemical parameter (except b*) were effectively predicted by methods of partial least squares regression (PLS), multiple linear regression (MLR) and principle component regression (PCR) using signals of electronic nose, electronic tongue. Electronic nose, electronic tongue and their fusion data have the ability in discrimination of mutton samples with different freeze-thaw cycles and in prediction of Cycles of mutton freeze-thawed and the physicochemical parameter. What’s more, better results were obtained by combining of electronic nose and electronic tongue.Fast authentication of meat was conducted by electronic nose and electronic tongue on minced mutton adulterated with pork, chicken at levels of0%,20%,40%,60%,80%and100%by weight, respectively. For electronic nose, the false rate and numbers of misclassified samples were used as evaluation criterion to estimate feature extraction methods, Principle component analysis, loading analysis and stepwise linear discriminant analysis (step-LDA) employed to optimize the data matrix of electronic nose. The adulterated mutton samples were successfully discriminated by PCA and CDA using data set extracted by step-LDA with correct accuracy as high as99.17%. Effective predictive models for the pork, chicken content in minced mutton built by PLS, MLR and Back propagation neural network (BPNN) were obtained with R2>0.9, the prediction error within10%.For electronic tongue, the adulterated mutton samples were grouped according to their content of chicken with few misclassified samples by means of PCA, and better classification were found by CDA with more samples correctly clustered. Predictive model for the pork, chicken content in minced mutton built by PLS, MLR and LS-SVM were obtained the highest R2of0.99, the lowest prediction error within3%. Adulterated mutton could be discriminated and successfully predicted both by electronic nose and electronic tongue by PLS, MLR and LS-SVM, and better results were obtained by signals of electronic nose.Data fusion methods were explored for integration of smell and taste information of minced mutton adulterated with pork and chicken obtained by electronic nose and electronic tongue by methods of combination of signals of electronic nose and electronic tongue data, normalization, data extraction by step-LDA and principle component analysis. The accuracy rate of discrimination and precision of prediction were all improved by the fusion of electronic nose and electronic tongue data.Integration of smell and taste information made the combination use of electronic nose and electronic tongue generating overall and comprehensive results of mutton samples, improving the identification accuracy and the validity of the quantitative prediction model.更多还原