【作者】 张军；

【导师】 李小昱；

【作者基本信息】 华中农业大学 ， 农业机械化工程， 2008， 博士

【摘要】 新鲜度是鱼类或鱼类制品质量的一个重要指标,对产品最终质量十分重要。鱼类是一种不耐保藏的食品,极易发生腐败变质,食用不当易发生食物中毒,因此在卫生和食品安全检验中,鱼肉新鲜度的快速、准确检测,将对鱼类的保鲜、储臧和深加工有着重要的科学意义和应用价值。论文以生物阻抗技术和电子鼻嗅觉技术来研究淡水鱼新鲜度的快速检测方法。采用生物阻抗技术和虚拟仪器技术,以四电极测量方法和扫频激励方式,构建了淡水鱼鱼体复阻抗测量系统,以鲫鱼和鲢鱼为研究对象,研究了鱼体复阻抗谱的测量方法和技术,研究了储藏过程中鱼体复阻抗特性(阻抗和相位)的变化规律,确定了最优的测量参数和条件,建立了基于复阻抗特性的鱼体新鲜度评价指标,在此基础上用BP-ANN(后向反馈神经网络)建立了淡水鱼新鲜度的预测模型。以仿生嗅觉技术和虚拟仪器技术,构建了淡水鱼电子鼻测量系统,测量了不同新鲜度的鲢鱼鱼肉,采用PCA(主成分分析)优化了电子鼻试验参数,确定了电子鼻的最优测量条件,采用PCR(主成分回归)、PLS(偏最小二乘法)和ANN(人工神经网络)等方法进行了建模分析,确定了最优的模型识别方法,并优化了电子鼻传感器阵列。主要研究结果如下:1.采用生物阻抗技术在100Hz～100kHz内测量了储藏过程中鲫鱼和鲢鱼的复阻抗谱,试验结果表明,不同储藏时间的复阻抗谱呈现cole圆弧特性并区分明显,为采用鱼体复阻抗特性评价淡水鱼新鲜度提供了理论依据。2.首次提出和建立了淡水鱼鱼体复阻抗特性的测量方法和测量系统,测量系统包括自行研制的四电极探头、压控恒流源电路、信号调理滤波电路等硬件部分和基于虚拟仪器的信号发生器、信号采集模块、信号处理和文件保存模块等软件部分。试验结果表明,该系统智能化程度高,能自动实现扫频测量和结果分析,并具有较高的测量精度,在100Hz～100kHz范围内阻抗幅值的最大误差为0.58%,频率在50kHz～100kHz相位最大误差为0.31°,而在频率小于50kHz的范围内,系统的相位的最大误差仅为0.07°;在100Hz～100kHz范围内,每倍频扫描10点,测量系统可在30～50s内获得被测鱼体组织阻抗幅值、相位随激励频率的变化而发生的变化,为进一步研究其它生物组织复阻抗特性奠定了良好的理论基础和提供了必备的技术支撑。3.针对生物阻抗测量中恒流源性能直接关系到测量精度,构建了基于AD844和AD620两种方案的压控恒流源,试验研究了两种方案的VCCS恒流源内阻和频率特性。试验结果表明,AD844构成的恒流源性能更好,其输出内阻为1155kΩ,大于AD620构成的VCCS输出阻抗500kΩ,确定了基于AD844构建的压控电流源性能最优,在此基础上增加了直流反馈电路,以保证电流源的稳定性。针对生物组织复阻抗特性相位角难以准确提取的特点,设计了5种相位差算法,通过对含有白噪声的信号进行试验分析,试验结果表明,数字解调法(加标准正余弦方法)的精度最高,相位最大误差小于0.03°。4.以鲫鱼和鲢鱼为研究对象,首次采用生物阻抗技术系统地研究了鱼体储藏过程中激励电流、测量电极、激励频率、测量方向和测量部位对鱼体复阻抗特性的影响,确定了淡水鱼鱼体复阻抗特性的最优测量条件是激励电流为0.3mA,测量电极为混合式电极,激励频率范围为1kHz～5kHz,测量部位为沿鳃部平行侧线方向。试验结果表明,不同部位的鱼体阻抗幅值大小不同,其大小顺序是尾部、腹部以及鳃部,其差异随储藏时间的增加而逐渐减小;在同一储藏时间下,鱼体阻抗幅值随激励频率的增加而减小,相位随激励频率的增加而增加,符合生物组织的频散特性;沿侧线平行方向测量的鱼体阻抗幅值比沿侧线方垂直向测量的结果要大,符合生物组织各向异性的特点,各向异性随储藏时间和激励频率的增加而逐渐减小,但不影响采用复阻抗特性评价淡水鱼新鲜度的指标,当沿侧线平行方向测量阻抗幅值与沿侧线垂直方向测量阻抗幅值的比值小于1.05时,均可认为鲫鱼或鲢鱼已经腐败。5.以TVB-N(挥发性盐基氮)作为新鲜度评价对照指标,在1kHz激励频率下确定了鲫鱼和鲢鱼新鲜度的复阻抗特性判定指标,并采用人工神经网络建立了最优预测模型。当鲫鱼鱼体阻抗幅值达到129.79Ω或相位达到5.43°,鲢鱼鱼体阻抗幅值达到91.26Ω或相位达到1.48°,均可判定鲫鱼或鲢鱼腐败。以阻抗幅值和相位为输入,TVB-N为输出,确定鲫鱼最优神经网络结构为2-11-1型,预测TVB-N与实测TVB-N的相关系数为0.95621,模型检验的准确率为94.44%;确定鲢鱼最优神经网络结构为2-10-1型,预测TVB-N与实测TVB-N的相关系数为0.99282,模型检验的准确率为97.44%。6.首次提出和构建了基于仿生嗅觉技术和虚拟仪器技术的淡水鱼电子鼻测量系统,测量系统包括电子鼻容器、4气体传感器阵列(TGS822乙醇类及有机溶剂气体型、TGS825硫化氢型、TGS826氨气及氨类型和TGS832卤烃型气体传感器)、调理电路等硬件部分与基于虚拟仪器的数据采集、信号分析和文件保存模块等软件部分。试验结果表明,电子鼻的响应随鱼肉新鲜度的变化而变化。采样方法将传统的静态顶空生成法改进成了自由扩散的顶空方法,简化了取样设备,为进一步研究便携式淡水鱼电子鼻测量系统奠定了理论基础和提供了技术支撑7.以鲢鱼为研究对象,采用PCA(主成分)分析方法,确定了顶空时间为5min、样品重量为60g为电子鼻评价鲢鱼新鲜度的最优测量条件。8.通过试验研究,确定了人工神经网络为电子鼻模型识别的最优建模方法,确定了TGS822,TGS826和TGS832为最优淡水鱼电子鼻传感器阵列组合。采用PCR(主成分回归)、PLS(偏最小二乘法)和BP-ANN(后向反馈人工神经网络)进行建模分析,各模型预测结果与实测TVB-N的最优相关系数为分别为0.89778,0.92250和0.97613,试验结果表明,采用BP-ANN建模精度最高,其最优网络结构为4-11-1型;在此基础上进一步采用ANN建模方法对传感器阵列进行优化,试验结果表明采用3传感器(TGS822,TGS826和TGS832)和原有4传感器测量结果的BP-ANN建模预测结果与实测TVB-N的相关系数均为0.976。9.最优3气体传感器阵列所建立的淡水鱼新鲜度等级预测模型中,人工神经网络网络模型结构为3-11-2时,模型对新鲜、次新鲜、半腐败以及腐败的预测准确率为97.32%,当采用复阻抗特性对电子鼻误判结果进行辅助判断,可将电子鼻的准确率提高到100%。论文创新之处在于:1.论文提出采用医学领域的生物阻抗技术进行淡水鱼新鲜度的检测,并验证了生物阻抗技术用于检测淡水鱼新鲜的的可行性,在此基础上开发了基于虚拟仪器的淡水鱼复阻抗测量系统,测量系统在100Hz～100kHz范围内阻抗幅值的最大误差为0.58%,在50kHz～100kHz相位最大误差为0.31°,而在频率小于50kHz的范围内,系统的相位的最大误差仅为0.07°,具有较高的测量精度。2.采用自行研制的淡水鱼鱼体复阻抗测量系统进行了大量的试验研究,确定了激励电流、测量电极、激励频率、测量部位和测量方向对鱼体复阻抗特性的影响,确定了混合电极电极、鱼体鳃部位和0.3mA的激励电流为最优测量条件,建立了评价淡水鱼新鲜度的复阻抗特性指标,在此基础上采用BP-ANN(后向反馈人工神经网络)建立了预测模型,对鲫鱼和鲢鱼新鲜度的预测准确率分别是94.44%和97.44%,提高了识别准确率。该研究建立了用复阻抗特性评价淡水鱼新鲜度的技术和方法,为评价淡水鱼新鲜度提供了一新的快速无损检测方法。3.将电子鼻的采样方法由传统的顶空法改进为自由扩散法,简化了取样设备,以虚拟仪器技术和仿生嗅觉技术为基础,首次提出和建立了4气体传感器阵列的淡水鱼电子鼻测量系统,并试验验证了电子鼻测量系统的可行性。系统地试验研究了影响电子鼻评价鱼体新鲜度的测量因素,确定了顶空5min和样品重量为60g为最优测量条件。采用BP-ANN(后向反馈人工神经网络)建立的淡水鱼新鲜度等级预测模型对新鲜、次新鲜、腐败以及半腐败的预测准确率为97.32%,当采用复阻抗特性进行辅助判断,可将电子鼻的准确率提高到100%,该项研究建立了用气体传感器评价淡水鱼新鲜度的技术和方法,为评价淡水鱼新鲜度提供了一新的快速检测方法。4.通过试验研究,将PCR(主成分回归)、PLS(偏最小二乘法)和BP-ANN(后向反馈人工神经网络)等方法用于电子鼻模式识别和建模分析,系统地探讨了不同建模方法对预测结果的影响,确定BP-ANN为最优建模方法。在此基础上,采用以上建模方法对气体传感器阵列进行优化,确定了TGS822,TGS826和TGS832为最优电子鼻气体传感器阵列,为电子鼻的传感器阵列优化提供了一种新方法。更多还原

【Abstract】 Freshness makes a major contribution to the quality of fish and fishery products. While,fish is a product of preservation intolerance and is easily deteriorated,the spoilage fish would lead to food poisoning.It is important to enhance hygienic examination to ensure the safety of fish products.Therefore,fast and accurate evaluation of fish freshness has a scientific and practical value to preservation,deposit and further processing of fish.Bioimpedance and electronic nose measurement technology was employed to research on fast evaluation freshness of freshwater fish.Based on bioimpedance and virtual instrument measurement technology,based on four-electrode and sweep frequency method,a freshwater fish bioimpedance measurement system was established.Crucian and silver carp was taken as study object,measurement method of bioimpedance spectrum and change rule of bioimepdance of freshwater fish during storage was investigated,the optimal measurement parameters and condition for the bioimpedance measurement of freshwater fish was ascertained.Futhermore,freshness index was constructed based on freshwater fish bioimpedance characteristics and BP-ANN （back-propagation artificial neural networks） method was used to improve the evaluation accuracy.A freshwater fish electronic nose measurement system was constructed based on artificial olfactory system and virtual instrument technology.Then,silver carp meat during storage was investigated by electronic nose measurement system.Experiment parameters of the system was optimized by PCA（Principal Component Analysis） method, and the optimal measurement condition was ascertained,finally,PCR（Principal Component Regression）,PLS（Partial Least Square） and ANN（Artificial Neural Network） methods were employed and compared to build mathematic model,the most optimal method of model identification was ascertained,furthermore,the gas sensor array was optimized by model comparison.The main conclusions are as follows.1.The bioimpedance spectrum of crucian and silver carp during storage presents as cole circle and the cole-cole plots measured at different post-mortem time differentiated obviously,the result provides a theoretical foundation for using bioimpedance technology and bioimpedance characteristics to evaluate freshness of freshwater fish.2.Measurement method and measurement system of bioimpedance characteristics of freshwater fish was firstly brought forward and constructed.The system consists of two parts,one is the hardware which includes self-design four-electrode,voltage control current source（VCCS）,signal processed circuit,and the other is software which includes signal generator,data acquisition and file auto-save model based on LabVIEW development environment.The result indicated that the system has high intelligence and accuracy,at the range of 100Hz～100 kHz,the maximum error of impedance is 0.58%,at the range of 50kHz～100kHz,the maximum error of phase is 0.31°;While at the frequency range lower than 50 kHz,the maximum error of phase is 0.07°.At the range of 100Hz～100kHz for 10 points every decade,it cost only 30s～50s for the system to calculate the impedance and phase changing with frequency,which establishes a theoretical foundation and technical parameters for other bio-tissue research on bioimpedance characteristics.3.Two VCCS based on AD844 and AD620 were designed to solve the problem of measurement accuracy of bioimpedance;the result indicated that the performance of VCCS based on AD844 is better,which has 1155 kΩfor inner impedance,while the inner impedance of VCCS based on AD620 is only 500 kΩ,in additional,a DCFB（Directed Current Feed Back） circuit was added to the VCCS to ensure the stability of current source.Aimed at solving the problem of phase accuracy of bio-tissue bioimpedance measurement,five different phase measurement methods was designed and then compared at the condition of white noise,the result indicated that the digital demodulation method（add standard sine and cosine method） has the highest accuracy,the maximum error is lower than 0.03°.4.The effect of excitation current,measurement electrode,excitation frequency, measured direction and measured position of crucian and silver carp during fish storage was first systematically investigated based on bioimpedance technology,the optimal measurement condition for freshwater fish bioimpedance measurement are as follows: 0.3mA for excitation current,composite electrode for measurement electrode,1kHz～5kH for excitation frequency and measured parallel to the lateral line of gill for measured direction and position.The result indicated that different measured positions lead to different impedance,the biggest impedance of measured position is tail,and then is belly and gill,however,the difference of impedance of measured position decreases with post-mortem time.At the same post-mortem time,the impedance decreases with the increase of frequency,while phase increases with frequency.The impedance changing rule conforms to the frequency dispersion of bio-tissue.At the same post-mortem time, the impedance measured paralleled to the lateral line is bigger than the impedance measured vertical to the lateral line of fish,which conforms to the anisotropy characteristics of bio-tissue,however,the anisotropy decreases with post-mortem time, when the ration of impedance measured parallel and vertical to the lateral line fish is lower than 1.05,it can indicate the deterioration of freshwater fish.5.Taking the TVB-N（Total Volatile Basic Nitrogen） as the compared index of freshwater fish freshness,the freshness index of bioimpedance characteristics of crucian and silver carp was constructed at 1 kHz and ANN method was used to build predicted model.When the impedance of crucian is less than 129.79Ωand phase of crucian is less than 5.43°,and when the impedance of silver carp is less than 91.26Ωand phase of silver carp is less than 1.48°,it can indicate the deterioration of freshwater fish.Taking the impedance and phase as input parameters and the TVB-N value as the output parameter for building the ANN predicted model,the result indicated that the optimal net layer structure for crucian ANN is 2-11-1,the model’s correlation coefficient of predicted and measured TVB-N is 0.95621,the test accuracy of the modle is 94.44%;while for silver carp,the optimal net layer structure is 2-10-1,the model’s correlation coefficient of predicted and measured TVB-N is 0.99282,the test accuracy of the model is 97.44%.6.A freshwater fish electronic nose was firstly brought forward and constructed based on artificial olfactory system and virtual instrument method,the measurement system includes electronic nose gas collection container,four-gas sensor array（TGS822 for ethanol;TGS825 for H₂S;TGGS826 for NH₃ and aime,TGS832 for halocarbon） and its signal processed circuit for hardware,and data acquisition,data analysis and file saving model based on LabVIEW for software.The result indicated that the response of electronic nose correlated with freshness loss of fish.A new air sample method（free air diffused method） was improved on the basis of traditional static headspace air analysis, which can simplify the sample equipment and provide theoretical foundation and technical parameters for constructing easy-carried electronic nose.7.Taking silver carp as the research object and PCA as the analysis method,the optimal measurement condition for electronic nose are as follows:headspace for 5min, 60g for sample weight.8.ANN was the most optimal model identification method for electronic nose,the three gas sensors,TGS822,TGS826 and TGS832,were the optimal gas sensor array for electronic nose.PCR,PLS and BP-ANN was used to build the predicted model,the result indicated their correlation coefficient of predicted and measured-N value are 0.89778, 0.92250 and 0.97613,which indicated that the ANN method has higher accuracy and is the optimal method,the optimal net layer of ANN structure is 4-11-1.On this basis,gas sensor array response were analyzed,the result showed that the three gas sensor（TGS822,TGS826 and TGS832） has the same correlation coefficient of predicted and measured-N value with the four gas sensor array（TGS822,TGS 825,TGS826 and TGS832）,the correlation coefficient is 0.976.The three gas sensors were ascertained as the optimal gas sensor array.9.The optimal net layer structure is 3-11-2,the accuracy of ANN model of optimal three gas sensor array is 97.32%to identify fresh,semi-fresh,semi-deteriorated and deteriorated fish samples.When using the bioimpedance index for assistant judgement of the freshness of fish,the accuracy can be improved to 100%.Innovations:1.Bioimpedance measurement technology was firstly brought forward to evaluate the freshness of freshwater fish and was proved to be able to evaluate the freshness of freshwater fish.On this basis,biomimpedance measurement system was constructed based on LabVIEW environment and has high accuracy.The maximum error of impedance of the system is 0.58%;the maximum error of phase is 0.31°and 0.07°during 50 kHz～100 kHz and 100 Hz～50 kHz.2.A large numbers of crucian and silver carp was used to investigate the factors affecting on bioimpedance characteristics,and the optimum measurement conditions were ascertained as follows:0.3mA for excitation current,composite electrode for measurement electrode,1 kHz～5 kHz for excitation frequency and measured parallel to the lateral line of gill for measured direction and position.Freshness index was constructed based on bioimpedance characteristics and BP-ANN method was used to construct evaluation freshness model to improve the accuracy,the model accuracy for crucian and silver carp is 94.44%and 97.44%.3.A new air sample method（free air diffused method） was improved on the basis of traditional static headspace air analysis,which simplified the sample equipment.Based on artificial olfactory technology and virtual instrument technology,electronic nose of freshwater fish was firstly brought forward and constructed with four gas sensor array and was proved validity.Experiment factors for electronic nose system was investigated, headspace for 5min and 60g for sample weight were ascertained as the optimum measurement conditions.On this basis,BP-ANN method was employed to build predicted model for freshness level,the accuracy is 97.32%to identify fresh,semi-fresh, semi-deteriorated and deteriorated fish samples.And when using the bioimpedance index for assistant judgement of the freshness of fish,the accuracy can be improved to 100%.4.PCR,PLS and BP-ANN was employed to build the predicted model and their model accuracy was compared,the result indicated that BP-ANN method was the optimum method for electronic nose system.On this basis,the gas sensor array was optimized by comparing the predicted accuracy,finally the three gas sensor（TGS822, TGS826 and TGS832） was ascertained as the optimum gas sensor array,which provides a ness method for optimizing gas sensor array.更多还原