基于近红外和高光谱检测鸡蛋粉掺假的研究
发布时间：2018-08-10 22:22
【摘要】：鸡蛋粉有高品质的蛋白质、均衡的矿物质和维生素等优良的营养特性,在食品工业和畜产工业中发挥着非常重要的作用。但是,有些商家为了降低生产成本向蛋粉中掺入廉价物质以获得更大的利润。所以,本文以鸡蛋粉(全蛋粉、蛋清粉、蛋黄粉)为研究对象,使用具有快速、无损检测等优点的近红外光谱技术和高光谱技术并结合有效的化学计量学方法对鸡蛋粉二元掺假体系和多元掺假体系进行了检测。同时采用两种光谱技术建立了全蛋粉中重要营养物质蛋白质和脂肪含量的检测模型,主要研究结果如下:(1)应用近红外光谱技术检测鸡蛋粉二元体系掺假。将掺伪物淀粉、大豆蛋白、麦芽糊精分别按比例掺入到三种鸡蛋粉中,构成鸡蛋粉的二元掺假体系。在三种鸡蛋粉的掺假检测中均建立定性判别模型(偏最小二乘判别模型)和定量检测模型(偏最小二乘回归模型)。结果显示,在三种鸡蛋粉掺假的判别分析中,偏最小二乘判别模型(PLS-DA)均可有效的将纯鸡蛋粉和掺假鸡蛋粉进行区分。在全蛋粉掺假的定量模型中,掺入淀粉、大豆蛋白和麦芽糊精的检测最佳模型均为回归系数-偏最小二乘回归模型(RC-PLSR),预测相关系数(Rp2)分别达到0.990、0.996和0.998;在蛋清粉掺假的定量模型中,掺入大豆蛋白和麦芽糊精的最佳检测模型为RC-PLSR,Rp2分别达到0.962和0.979,而对于掺入淀粉的蛋清粉的检测模型,经过回归系数法(RC)获得的特征波长所建立的RC-PLSR在性能上显著降低,因此选择全波段波长建立的PLS模型,获得的Rp2为0.921;在蛋黄粉掺假的定量模型中,掺入淀粉、大豆蛋白和麦芽糊精的RC-PLSR相较于最佳波段获得的PLS模型性能均有所降低,因此均选用最佳波段下获得的模型,Rp2分别达到0.998、0.997和0.986。(2)近红外光谱技术检测鸡蛋粉多元体系掺假。将淀粉、大豆蛋白和麦芽糊精三种掺杂物两两或者三者的混合物按比例掺入到鸡蛋粉中,构成鸡蛋粉的多元掺假体系。分别建立鸡蛋粉中淀粉、大豆蛋白、麦芽糊精掺伪量和总掺伪量的定量模型,并比较了不同预处理方式和不同波段对于各个模型的影响。基于最佳预处理方式和最优波段建立主成分回归(PCR)定量模型并与PLSR模型性能进行比较。结果发现,淀粉掺伪量模型和麦芽糊精掺伪量模型效果不佳,而大豆蛋白掺伪量和总掺伪量模型的预测性能良好,Rp2均达到0.950以上,PLSR模型检测性能优于PCR模型。(3)应用高光谱技术检测鸡蛋粉二元体系掺假。通过采集纯样品和掺假样品的高光谱图像并提取平均光谱,建立支持向量机(SVM)模型对纯蛋粉和掺假蛋粉进行判别,结果表明三种鸡蛋粉掺假的判别正确率都达到90%以上。为了定量检测掺入物的含量,采用偏最小二乘回归模型建立光谱数据与掺假含量之间的关系。结果显示全蛋粉掺假的研究中,掺入淀粉、大豆蛋白、麦芽糊精所建立的PLSR模型的Rp2分别达到0.931、0.981、0.990;蛋清粉掺假的研究中,掺入淀粉、大豆蛋白、麦芽糊精所建立的PLSR模型的Rp2分别达到0.832、0.994和0.984;蛋黄粉掺假的研究中,掺入淀粉、大豆蛋白、麦芽糊精所建立的PLSR模型的Rp2分别达到0.998、0.986和0.975,说明模型具有良好性能。通过RC法和连续投影法(SPA)提取了掺假的重要特征波长,分别建立RC-PLSR和SPA-PLSR简化模型。结果表明,模型在性能上没有显著差别,但是由于减少了波长的数量,使得运算的时间缩减、效率提高。(4)高光谱技术检测鸡蛋粉多元体系掺假。在掺假检测中,定性判别模型采用了随机森林(RF)和支持向量机(SVM)的方法,结果显示二者均能对纯蛋粉和掺假蛋粉进行有效判别,且RF的效果略优于SVM。在掺入混合物的全蛋粉、蛋清粉、蛋黄粉的定量模型中建立蛋粉实际含量与预测含量之间的关系,Rp2分别可以达到0.986、0.992、0.989,说明模型性能良好。为了简化模型,根据RC法和SPA法提取特征波长,分别建立了RC-PLSR、RC-MLR及SPA-PLSR、SPA-MLR模型,所建立的模型表现出了良好的性能。(5)采用近红外光谱和高光谱技术建立全蛋粉中蛋白质和脂肪含量的定量模型,并比较了两种方法的性能。在近红外光谱测定中,蛋白质检测的最优模型为回归系数-PLSR,Rp2达到0.996,脂肪检测的最优模型为基于全波段所建立的模型,Rp2达到0.974;在高光谱测定中,蛋白质检测的最优模型为载荷系数-PLSR,Rp2达到0.995,脂肪检测的最优模型为回归系数-PLSR,Rp2达到0.964。两种方法均表现出良好的性能,近红外光谱检测技术的性能优于高光谱。
[Abstract]:Egg powder has high quality protein, balanced mineral and vitamin nutritional properties and plays a very important role in the food industry and livestock industry. However, in order to reduce production costs, some businesses add cheap substances into egg powder to obtain greater profits. Egg yolk powder (Egg yolk powder) was studied by near infrared spectroscopy (NIRS) and hyperspectral techniques with the advantages of rapid and non-destructive detection, combined with effective chemometrics to detect the binary adulteration system and multicomponent adulteration system of egg powder. The main results are as follows: (1) Near-infrared spectroscopy was used to detect the adulteration of egg powder binary system. Adulterated starch, soybean protein and maltodextrin were mixed into three kinds of egg powder in proportion to form a binary adulteration system of egg powder. The results showed that the partial least squares discriminant model (PLS-DA) could distinguish the pure egg powder from the adulterated egg powder in the discriminant analysis of three kinds of egg powder adulteration. In the quantitative model of egg powder adulteration, starch and soybean protein were added. The best detection models for soybean protein and maltodextrin were both regression coefficient-partial least squares regression model (RC-PLSR), and the predictive correlation coefficient (Rp2) was 0.990, 0.996 and 0.998, respectively. In the quantitative model of egg white powder adulteration, the best detection models for soybean protein and maltodextrin were RC-PLSR, Rp2 was 0.962 and 0.979, respectively, and for starch adulteration. In the detection model of egg yolk powder, the characteristic wavelengths obtained by regression coefficient method (RC) significantly reduced the performance of RC-PLSR, so the PLS model with full-band wavelength was selected, and the Rp2 was 0.921. In the quantitative model of egg yolk powder adulteration, the RC-PLSR with starch, soybean protein and maltodextrin was obtained at the optimum band. The performances of PLS models were decreased, so the best band models were selected, and Rp2 was 0.998, 0.997 and 0.986 respectively. (2) Near infrared spectroscopy was used to detect the adulteration of egg powder. The quantitative models of starch, soybean protein, maltodextrin adulteration and total adulteration in egg powder were established, and the effects of different pretreatment methods and different wave bands on each model were compared. The results showed that starch adulteration model and maltodextrin adulteration model were not effective, while soybean protein adulteration and total adulteration model had good predictive performance, Rp2 was above 0.950, and PLSR model had better detection performance than PCR model. (3) Detection of egg powder adulteration by Hyperspectral technique. Supporter Vector Machine (SVM) model was established to discriminate pure egg powder and adulterated egg powder. The results showed that the correct discriminant rate of three kinds of egg powder adulteration was above 90%. In order to quantitatively detect the content of adulterated egg powder, partial least squares regression model was used to establish spectral data and adulterated egg powder. The results showed that the Rp2 of the PLSR models with starch, soybean protein and maltodextrin were 0.931, 0.981, 0.990 respectively, and that of the PLSR models with starch, soybean protein and maltodextrin were 0.832, 0.994 and 0.984 respectively. In the study, the Rp2 of PLSR model with starch, soybean protein and maltodextrin was 0.998, 0.986 and 0.975 respectively, which showed that the model had good performance. The important characteristic wavelengths of adulteration were extracted by RC method and continuous projection method (SPA), and the simplified models of RC-PLSR and SPA-PLSR were established respectively. No, but because the number of wavelengths is reduced, the calculation time is shortened and the efficiency is improved. (4) Hyperspectral technique is used to detect the adulteration of egg powder. In order to simplify the model, RC-PLSR, RC-M were established according to the characteristic wavelengths extracted by RC and SPA methods. LR, SPA-PLSR and SPA-MLR models showed good performance. (5) Quantitative models of protein and fat content in egg powder were established by near infrared spectroscopy and hyperspectral techniques, and the performances of the two methods were compared. 96, the optimal model for fat detection is based on the full-band model, Rp2 reaches 0.974; in hyperspectral measurement, the optimal model for protein detection is load coefficient-PLSR, Rp2 reaches 0.995, the optimal model for fat detection is regression coefficient-PLSR, Rp2 reaches 0.964. Both methods show good performance, and near-infrared spectroscopy detection technology. The performance of the technique is better than that of hyperspectral imaging.
【学位授予单位】：华中农业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：O657.3;TS253.7
[Abstract]:Egg powder has high quality protein, balanced mineral and vitamin nutritional properties and plays a very important role in the food industry and livestock industry. However, in order to reduce production costs, some businesses add cheap substances into egg powder to obtain greater profits. Egg yolk powder (Egg yolk powder) was studied by near infrared spectroscopy (NIRS) and hyperspectral techniques with the advantages of rapid and non-destructive detection, combined with effective chemometrics to detect the binary adulteration system and multicomponent adulteration system of egg powder. The main results are as follows: (1) Near-infrared spectroscopy was used to detect the adulteration of egg powder binary system. Adulterated starch, soybean protein and maltodextrin were mixed into three kinds of egg powder in proportion to form a binary adulteration system of egg powder. The results showed that the partial least squares discriminant model (PLS-DA) could distinguish the pure egg powder from the adulterated egg powder in the discriminant analysis of three kinds of egg powder adulteration. In the quantitative model of egg powder adulteration, starch and soybean protein were added. The best detection models for soybean protein and maltodextrin were both regression coefficient-partial least squares regression model (RC-PLSR), and the predictive correlation coefficient (Rp2) was 0.990, 0.996 and 0.998, respectively. In the quantitative model of egg white powder adulteration, the best detection models for soybean protein and maltodextrin were RC-PLSR, Rp2 was 0.962 and 0.979, respectively, and for starch adulteration. In the detection model of egg yolk powder, the characteristic wavelengths obtained by regression coefficient method (RC) significantly reduced the performance of RC-PLSR, so the PLS model with full-band wavelength was selected, and the Rp2 was 0.921. In the quantitative model of egg yolk powder adulteration, the RC-PLSR with starch, soybean protein and maltodextrin was obtained at the optimum band. The performances of PLS models were decreased, so the best band models were selected, and Rp2 was 0.998, 0.997 and 0.986 respectively. (2) Near infrared spectroscopy was used to detect the adulteration of egg powder. The quantitative models of starch, soybean protein, maltodextrin adulteration and total adulteration in egg powder were established, and the effects of different pretreatment methods and different wave bands on each model were compared. The results showed that starch adulteration model and maltodextrin adulteration model were not effective, while soybean protein adulteration and total adulteration model had good predictive performance, Rp2 was above 0.950, and PLSR model had better detection performance than PCR model. (3) Detection of egg powder adulteration by Hyperspectral technique. Supporter Vector Machine (SVM) model was established to discriminate pure egg powder and adulterated egg powder. The results showed that the correct discriminant rate of three kinds of egg powder adulteration was above 90%. In order to quantitatively detect the content of adulterated egg powder, partial least squares regression model was used to establish spectral data and adulterated egg powder. The results showed that the Rp2 of the PLSR models with starch, soybean protein and maltodextrin were 0.931, 0.981, 0.990 respectively, and that of the PLSR models with starch, soybean protein and maltodextrin were 0.832, 0.994 and 0.984 respectively. In the study, the Rp2 of PLSR model with starch, soybean protein and maltodextrin was 0.998, 0.986 and 0.975 respectively, which showed that the model had good performance. The important characteristic wavelengths of adulteration were extracted by RC method and continuous projection method (SPA), and the simplified models of RC-PLSR and SPA-PLSR were established respectively. No, but because the number of wavelengths is reduced, the calculation time is shortened and the efficiency is improved. (4) Hyperspectral technique is used to detect the adulteration of egg powder. In order to simplify the model, RC-PLSR, RC-M were established according to the characteristic wavelengths extracted by RC and SPA methods. LR, SPA-PLSR and SPA-MLR models showed good performance. (5) Quantitative models of protein and fat content in egg powder were established by near infrared spectroscopy and hyperspectral techniques, and the performances of the two methods were compared. 96, the optimal model for fat detection is based on the full-band model, Rp2 reaches 0.974; in hyperspectral measurement, the optimal model for protein detection is load coefficient-PLSR, Rp2 reaches 0.995, the optimal model for fat detection is regression coefficient-PLSR, Rp2 reaches 0.964. Both methods show good performance, and near-infrared spectroscopy detection technology. The performance of the technique is better than that of hyperspectral imaging.
【学位授予单位】：华中农业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：O657.3;TS253.7
【参考文献】
相关期刊论文 前10条
1 王文秀;彭彦昆;孙宏伟;王凡;田芳;陈兴海;;基于可见/近红外光谱生鲜肉多品质参数检测装置研发[J];农业工程学报;2016年23期
2 罗微;杜焱U，
本文编号：2176380
本文链接：https://www.wllwen.com/kejilunwen/huaxue/2176380.html
教材专著
