【作者】 樊玉霞；

【导师】 成芳；

【作者基本信息】 浙江大学 ， 农业机械化工程， 2011， 博士

【摘要】 肉及肉制品是人类获得蛋白质、维生素和矿物质等营养成分的重要来源之一,猪肉作为我国生产和消费最大的畜肉产品,已成为人们饮食结构的基本组成部分。随着人们生活水平的提高和膳食结构的改变,对猪肉品质与安全提出了更高的要求。我国作为猪肉生产消费大国,但猪肉产品却难以进入国际市场,猪肉品质检测技术和评价方法繁琐,降低了监管机构的工作效率,因此实现猪肉品质与安全的快速检测和评价已成为猪肉行业发展和保证食品安全的重要环节之一本研究以新鲜猪肉肉糜为研究对象,利用可见/近红外光谱分析技术、现代仪器分析技术和化学计量学方法,开展了猪肉营养品质和安全品质的快速检测研究,主要对猪肉肉糜中粗脂肪、肌内脂肪、蛋白质、水分和14种脂肪酸进行了定量检测研究,以及不同部位肌肉肉糜的定性鉴别分析研究；针对安全品质检测,.对猪肉肉糜掺假检测和新鲜度的检测进行了研究,建立了掺假和新鲜度评价方法,对不同建模方法、不同光谱预处理和不同仪器进行了比较分析。主要研究结论如下：(1)研究了新鲜猪肉肉糜品质指标的定量分析。采用便携式光谱仪对猪肉背最长肌样本的肌内脂肪、蛋白质和水分含量进行了定量检测,比较了SMLR、PLSR和LS-SVM三种定量方法,结果表明SMLR模型性能整体较差,PLSR和LS-SVM两种方法结果较好,原始光谱建立的3个指标的PLSR模型相对最优,OSC光谱预处理后建立的3个指标的LS-SVM模型相对较优,两种方法建立的肌内脂肪、蛋白质和水分含量的相对较优模型性能差异不大。进一步采用不同光谱仪对肌内脂肪含量的定量分析进行了研究,基于傅立叶变换近红外光谱所建PLSR和LS-SVM模型结果差异不大,基于便携式光谱仪采集可见/近红外光谱所建PLSR模型结果优于LS-SVM模型,傅立叶变换近红外光谱所建模型性能优于便携式光谱仪采集的可见/近红外光谱所建模型；相对最优PLSR模型的校正集和预测集相关系数分别为0.964、0.960,RMSEC和RMSEP分别为0.249和0.491,相对最优LS-SVM模型的校正集和预测集相关系数分别为0.966、0.960,RMSEC和RMSEP分别为0.249和0.444。为扩大品质指标含量的范围,对采集至猪肉不同部位肌肉的肉糜样本的品质组分定量分析进行了研究,比较了PLSR和LS-SVM两种定量校正方法对脂肪、蛋白质和水分含量的检测,对脂肪、蛋白质和水分含量的LS-SVM模型整体好于PLSR模型；对脂肪含量的分析基于傅立叶变换近红外光谱建立的模型优于基于便携式光谱仪采集光谱建立的模型,对蛋白质和水分含量的分析模型基于USB4000采集光谱建立的模型较优；综合分析脂肪、蛋白质和水分含量最优分析模型,脂肪和水分含量模型的校正及预测相关系数均高于0.9,模型的预测精度及稳定性均较好,但蛋白质含量模型校正及预测相关系数仅高于0.7,模型精度相对较差。(2)研究了不同部位猪肉肉糜的定性判别分析。主要对4种不同肌肉类型的定性分析,比较了不同仪器,不同定性分析方法和不同光谱预处理所建鉴别模型的性能。不同光谱预测对模型性能产生的影响,结果表明原始光谱或一阶微分光谱所建模型的性能较优；不同光谱采集仪器的对比结果表明,便携式光谱仪获得的可见/近红外光谱更适用于不同肌肉类型猪肉肉糜的定性鉴别分析；比较不同的定性分析方法,DA方法所建模型性能略差,PLSDA最优模型校正及预测判别正确率分别为100%和94%,LA-SVMDA最优模型校正及预测判别正确率分别为99%和98%,表明可见/近红外光谱分析可快速鉴别不同部位猪肉肉糜。(3)对猪肉肉糜脂肪酸含量进行了定量检测研究。利用可见/近红外光谱分析技术对猪肉背最长肌中14种脂肪酸含量的定量预测进行了分析,主要比较了不同便携式光谱仪获得的不同可见/近红外光谱建模、PLSR和LS-SVM两种不同建模方法及不同光谱预处理所建立的模型结果,两种不同的光谱仪和两种不同的定量分析方法对脂肪酸含量的检测没有显示出明显的优势,只是对某些指标的检测结果有相对较好的预测结果：比较C14：0、C16:0、C16:1、C17:0、C18:0、C18:1、C18:2、C18:3、C20:1、C20:4、C20:5、SFA、MUFA和PUFA的相对较优定量模型,其中C 17：0、C18：3和SFA模型相关性、模型精度及稳定性较优,C20：4、C20：5和MUFA模型性能性对略差,各指标检测精度需要进一步提高。(4)对猪肉肉糜掺假检测进行了研究。采用傅立叶变换近红外光谱仪和便携式光谱仪USB4000获得同源掺假和非同源掺假两类样本不同等级掺假的可见/近红外光谱,建立不同等级掺假的定量分析模型。比较PLSR、PCR和MLR三种定量分析方法对不同等级掺假检测的建模结果,模型的总体性能差异不大,考虑到模型的稳定性、最优精度和适应性,比较不同波段、不同方法、不同掺假类型的模型性能,便携式光谱仪的检测结果较好,基于多元线性回归模型的性能较好,对同源掺假检测,基于可见/近红外光谱的MLR最优模型,校正预测及交互验证相关系数r分别为0.965、0.958和0.949,RMSEC、RMSEP和RMSECV分别为0.083、0.092和0.100；对非同源掺假检测,基于可见/近红外波段的MLR模型性能最优,校正预测及交互验证相关系数r分别为0.961、0.971和0.945,RMSEC、RMSEP和RMSECV分别为0.087、0.078和0.104,模型最稳定,预测精度最好,适用于猪肉肉糜掺假的快速检测。(5)研究了猪肉新鲜度的定量评价方法。利用便携式光谱仪扫描采集保存不同时间猪肉的可见/近红外光谱,采用SMLR, PCR和PLSR三种定量分析方法建立新鲜度的预测模型,比较不同预处理对模型性能的影响,SMLR模型总体性能最差,PCR模型总体性能次之,PLSR模型的总体性能较优,3点光谱建立的PLSR模型结果最优,校正预测及交互验证相关系数r分别为0.970、0.956和0.887,RMSEC、RMSEP和RMSECV分别为2.63 mg/100g、3.45 mg/100g和5.01mg/100g。研究结果表明,可见/近红外光谱分析技术能实现对猪肉新鲜度的快速检测,对实现监测猪肉品质安全提供了保障和依据。更多还原

【Abstract】 Meat and meat products are one of the essential sources of human nutritional component, such as protein, vitamins and minerals. Pork is one of the main components of human diet, which is the largest production and consumption in China. Along with improvement of living standard and changing of meal structure, the demand for pork quality and safety should be higher than before. China is a major pork producer as well as a major pork consumer, but not powerful in international market. Testing technology and assessment method have complicated formalities, which reduce the efficiency of regulatory authority. Therefore, rapid detection and evaluation of pork quality and safety become one of important part in the development of pork industry and the assurance of food safety.The research object is fresh minced pork. Detection of pork quality and safety were carried out based on visible/near infrared (NIR) spectroscopy, chemometrics techniques combined with modern physical-chemical analysis techniques. Quantitative models were established based on visible/NIR spectra for fat, intramuscular fat, protein, moisture and fatty acids determination, and qualitative models were also established for muscle discriminant from different part of pig. In this dissertation, detection of adulteration and freshness using visible/NIR spectra were also studied for safe quality detection of pork.The main results and conclusions were:(1) Quantitative analysis of minced pork quality was studied. Intramuscular fat, protein, moisture content in minced pork form longissimus dorsi muscle were analyzed quantitatively based on visible/NIR spectra by a portable instrument. Comparison results of different calibration methods of stepwise multi linear regression (SMLR), partial least squares regression (PLSR), least squares support vector machines (LS-SVM) and different spectra pretreatments showed that the performance of PLSR and LS-SVM model was much better than SMLR model. The best model of PLSR was established based on the raw spectra, and the best models of LS-SVM was based on the spectra with orthogonal signals correction (OSC) pretreatment for intramuscular fat, protein and moisture content, which was closed to PLSR calibration performance. Further research for quantitative analysis of Intramuscular fat was studied. Comparison results of different calibration methods of PLSR, LS-SVM and different spectra pretreatments and different detectors showed that the performance of model based on the spectra acquired by InGaAs detector was much better. The performance of LS-SVM model was close to that of PLSR model, but the performance of PLSR model based on the spectra acquired by USB4000 was much better than the LS-SVM model. For the best PLSR model, the correlation coefficients of calibration and validation were 0.964 and 0.960, respectively; the root mean square error of calibration (RMSEC) and root mean square error of prediction (RMSEP) were 0.249and 0.491, respectively. For the best LS-SVM model, the correlation coefficients of calibration and validation were 0.966 and 0.960, respectively; RMSEC and RMSEP were 0.249 and 0.444, respectively. In order to expanding the range of the content for quality index, the samples of minced pork collected from different muscle of pig were analyzed quantitatively. Comparison results of different calibration methods of PLSR, LS-SVM and different spectra pretreatments and different detectors showed that the performance of LS-SVM model was better than PLSR model for fat, protein and moisture content. For fat, the model based on the spectra acquired by InGaAs detector performed better. For protein and moisture, the model based on the spectra acquired by USB4000 performed better. The correlation coefficients for calibration and validation were above 0.9 for detection of fat and moisture, and the precision and stability were strong. But for the protein, the correlation coefficient for calibration and validation were above 0.7, the precision was relatively worse.(2) Qualitative analysis of minced pork from 4 different muscle was studied. Comparison results of different calibration methods of discriminant analysis (DA), partial least squares discriminant analysis (PLSDA), least squares support vector machines discriminant analysis (LS-SVMDA) and different spectra pretreatments and different detectors showed that the model based on the raw spectra or first derivative performed better. Visible/NIR spectra obtained by portable instrument were more suitable for discrimination of 4 different muscles. The discriminant results of DA model were a bit worse than PLSDA model and LA-SVMDA model. The discriminant accuracy of calibration and validation for the best PLSDA model were 100% and 94%, respectively and for LA-SVMDA were 99% and 98%, respectively.(3) Quantitative analysis of fatty acids was researched. Comparison results of different calibration methods of PLSR, LS-SVM and different spectra pretreatments and different detectors for detection of fatty acid content showed that there were no significant difference between PLSR and LS-SVM or two different detectors as a whole. The correlations, precision and stability were relatively better for C17:0, C18:3 and SFA, but worse for C20:4, C20:5 and MUFA. The precision of quantitative analysis for fatty acids should be improved.(4) Quantitative analysis of adulteration in minced pork were studied with different grade adulteration based on different detectors. Comparison results of different calibration methods and different wavebands and different type adulteration showed that the performance of multi linear regression (MLR) model based on visible/NIR spectra was much better, the correlation coefficients of calibration, validation and cross validation were 0.965,0.958 and 0.949, respectively; RMSEC, RMSEP and RMSECV were 0.083,0.092 and 0.100, respectively for adulterated with the same homologous material, and the correlation coefficients of calibration, validation and cross validation were 0.961,0.971 and 0.945, respectively; RMSEC, RMSEP and RMSECV were 0.087,0.078 and 0.104, respectively for adulterated with the nonhomologous material.(5) Quantitive analysis of freshness of pork were studied. The models established using SMLR, PCR and PLSR based on the visible NIR spectra. Comparison results of different calibration methods and different spectra pretreatment showed that the performance of PLSR model was much better and the SMLR model was much worse. The best model using first derivative spectra achieved the correlation coefficients of calibration, validation and cross validation were 0.970,0.956 and 0.887, respectively; RMSEC, RMSEP and RMSECV were 2.63 mg/100g,3.45 mg/100g and 5.01 mg/100g, respectively. This study demonstrated that visible/NIR spectroscopy can be successfully applied as a rapid method to determine the quality and safety of pork.更多还原