基于显微高光谱成像技术的滩羊肉品质检测研究

发布时间：2019-04-27 03:22

【摘要】：本文自行设计搭建一个显微高光谱成像系统,其融合了高光谱成像技术和显微成像技术,通过对滩羊肉样本光谱成像,获取样本的显微图像及光谱信息,初步研究了贮藏过程中羊肉组织结构变化,为羊肉贮藏过程中品质变化机理的研究提供理论依据。主要研究内容如下:(1)系统搭建及优化:以分立单元成像光谱仪、显微镜、数据采集卡等搭建显微高光谱成像系统,分析显微高光谱成像系统的成像原理。对系统的关键技术进行了研究,给出系统的技术指标。最后,对显微高光谱成像系统进行优化。(2)对羊肉贮藏过程中的pH、肉色、菌落总数、TVB-N含量和水分含量的变化规律进行了研究,并对各品质指标与贮藏时间及各品质指标间的相关性进行了分析,结果表明:水分含量、菌落总数和TVB-N含量与冷藏时间极显著相关(p0.01),相关系数分别为-0.992、0.995、0.991。进一步探讨了水分含量、菌落总数和TVB-N含量与冷藏时间之间的关系,建立水分含量、菌落总数和TVB-N含量与冷藏时间之间的曲线回归模型,进行拟合分析。得到回归方程分别为 Y=-2.604X2+0.064X+68.623,Y=0.179X2+0.015X+4.359,Y=1.031X2+0.108X+7.448。(3)以羊肉为研究对象,以贮藏过程中羊肉品质指标水分含量、菌落总数和TVB-N含量为评价指标,采用4种不同的光谱预处理方法进行光谱预处理优选最佳光谱预处理方法,最后结合不同的建模方法分别建立水分含量、羊肉菌落总数和TVB-N含量与冷藏时间的预测模型,优选最佳模型。结果显示:光谱数据经过正交信号校正后的光谱建立水分含量、菌落总数和TVB-N含量的预测模型效果较好,其Rc分别为0.9426、0.9696和0.9695,RP分别为0.9122、0.9201和0.9069高于其他光谱预处理模型。通过不同建模方法的比较,建模效果较好的是PLSR方法,其Rc分别为0.9195、0.9067和0.9147,Rp分别为0.8795、0.8743和0.8802,均优于PCR和SVR模型。因此,采用高光谱成像技术可实现羊肉品质指标的定量分析。(4)对羊肉贮藏过程中组织结构变化进行分析研究。首先获取羊肉样本的显微高光谱图像,并结合显微镜对羊肉不同贮藏时间的显微结构图进行观察分析;通过主成分分析法对图像进行降维处理,筛选617nm、622nm、632nm、767nm、875nm和966nm六个波长,作为特征波长;对这些特征波长下的显微图像进行分析,发现羊肉组织结构随着贮藏天数的增加,破坏程度也增加。研究结果表明:运用显微高光谱成像技术,可以对羊肉贮藏过程中的组织结构变化进行分析。本研究采用菌落总数对羊肉新鲜度进行表征,提取羊肉显微高光谱图像信息的纹理特征,运用SVM和LDA两种方法对羊肉的新鲜度等级进行划分,其校正集判别率分别为98.33%和91.67%,预测集判别率分别为93.33%、93.33%,SVM法判别效果较好。因此,显微高光谱成像技术结合适合的算法,可实现羊肉贮藏过程中新鲜度等级分类判别,为羊肉贮藏过程中的品质变化机理研究奠定了基础。
[Abstract]:In this paper, a micro-hyperspectral imaging system is designed and built, which combines hyperspectral imaging technology with microscopic imaging technology. Through spectral imaging of mutton samples, the microscopic images and spectral information of the samples are obtained. The changes of tissue structure of mutton during storage were studied, which provided theoretical basis for studying the mechanism of mutton quality change during storage. The main contents are as follows: (1) system construction and optimization: the micro-hyperspectral imaging system is constructed by discrete unit imaging spectrometer, microscope, data acquisition card and so on, and the imaging principle of micro-hyperspectral imaging system is analyzed. The key technology of the system is studied and the technical index of the system is given. Finally, the microscopic hyperspectral imaging system was optimized. (2) the changes of pH, color, total colony count, TVB-N content and water content of mutton during storage were studied. The correlation of each quality index with storage time and quality index was analyzed. The results showed that water content, total colony count and TVB-N content were significantly correlated with cold storage time (p0.01), and the results showed that water content, colony count and TVB-N content were significantly correlated with cold storage time (p0.01). The correlation coefficients were-0.992,0.995,0.991.The correlation coefficients were-0.992,0.995,0.991. Furthermore, the relationship among water content, total colony count and TVB-N content and storage time was discussed. The curve regression model of water content, total colony count and TVB-N content and cold storage time was established, and the fitting analysis was carried out. The regression equations were Y=-2.604X2 0.064X 68.623, Y = 0.179X2 0.015X 4.359, Y = 1.031X2 0.108X 7.448. (3) the water content of mutton quality index during storage was studied, and the regression equation was Y=-2.604X2 0.064X 68.623, Yx0.179X2 0.015X 4.359, Y = 1.031X2 0.108X 7.448 respectively. The total number of colonies and the content of TVB-N were the evaluation indexes. Four different spectral pretreatment methods were used to optimize the optimum spectral pretreatment methods. Finally, the water content was established by combining different modeling methods. The best model was selected to predict the total colony count, TVB-N content and cold storage time of mutton. The results showed that after the spectral data were corrected by orthogonal signal, the prediction model of total colony count and TVB-N content was better, and its Rc were 0.9426,0.9696 and 0.9695, respectively. The RP values were 0.9122, 0.9201 and 0.9069, respectively, which were higher than those of other spectral pretreatment models. Through the comparison of different modeling methods, the better modeling effect is PLSR method, whose Rc is 0.9195, 0.9067 and 0.9147, respectively, and RP is 0.8795, 0.8743 and 0.8802, which are better than PCR and SVR model. Therefore, the quantitative analysis of mutton quality indexes can be achieved by using hyperspectral imaging technique. (4) the changes of tissue structure of mutton during storage were analyzed and studied. Firstly, the microscopic hyperspectral images of mutton samples were obtained, and the microstructure of mutton at different storage times was observed and analyzed with microscope. Six wavelengths of 617 nm, 622 nm, 632 nm, 767 nm, 875 nm and 966 nm were selected by principal component analysis. By analyzing the microscopic images at these characteristic wavelengths, it was found that the damage degree of mutton tissue structure increased with the increase of storage days. The results show that the microstructure changes of mutton during storage can be analyzed by micro-hyperspectral imaging technique. In this study, the freshness of mutton was characterized by the total number of colonies, the texture characteristics of mutton micro-hyperspectral image were extracted, and the grade of freshness of mutton was classified by SVM and LDA. The calibration set discrimination rate is 98.33% and 91.67% respectively, and the predictive set discrimination rate is 93.33% and 93.33%, respectively. SVM method has a better discriminant effect. Therefore, micro-hyperspectral imaging combined with suitable algorithms can be used to classify and distinguish freshness of mutton during storage, which lays a foundation for studying the mechanism of mutton quality change during storage.
【学位授予单位】：宁夏大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TS251.53;O657.3

【参考文献】