【作者】 马飞；

【导师】 郑磊；

【作者基本信息】 合肥工业大学 ， 食品科学， 2015， 博士

【摘要】 肉制品是居民饮食的重要组成部分,其品质优劣不仅影响消费者的健康状况,而且也决定了肉类行业的发展。因此,如何进行肉制品品质检测就显得十分重要,也是食品科学研究的热点之一。目前,肉制品品质评价主要由感官、理化及微生物学等传统检测方法来完成,但存在检测效率低、破坏性大和检测成本高等一系列缺点,无法满足现代工业对肉制品品质实时监控的要求,因而寻找快速、无损的肉制品品质检测方法变得非常迫切。本文以不同加工和贮藏方式的香肠为研究对象,利用波长范围为405～970nm的多光谱成像(MSI)系统获取样品信息,采用偏最小二乘回归(PLSR)、支持向量机(SVM)、主成分分析(PCA)、连续投影算法(SPA)和灰度共生矩阵(GLCM)等方法对信息进行分析,探求对香肠多元品质特征的无损检测。完成的主要研究结果如下：1)基于光谱信息,SVM和PLSR可分别建立精确的香肠硬度和菌落总数预测模型,其中硬度模型的决定系数(RP2)和相对预测误差(RPD)分别为0.546和1.455,菌落总数模型的RP2和RPD分别为0.891和2.970。2)将主成分图像的纹理信息与光谱信息组合可提高香肠水分含量、保水性、凝聚性、咀嚼性、血红素铁和非血红素铁的检测精度,获得最佳预测模型,其RP2分别为0.899、0.691、0.619、0.728、0.912和0.901,相应的RPD分别为3.027、1.800、1.561、1.770、3.356和3.167；另外,最佳模型的精确度与主成分的贡献率无关。3)利用图像处理和模型插值运算成功得出香肠水分含量、保水性、咀嚼性、血红素铁、非血红素铁和菌落总数的可视化分布图,由此可直观地评价香肠的品质变化,充分体现出MSI的优越性。本研究表明MSI技术可精确地对香肠多元品质(除弹性)进行快速、无损检测,满足肉类工业对香肠加工和贮藏过程中的品质实时监控；研究结果可为MSI技术在肉类工业上的应用提供理论参考,也充分体现出MSI技术的重要研究价值。更多还原

【Abstract】 Meat product plays an important role in our diet, and the quality characteristics of which not only affect the health of consumers, but also determine the development of meat industry. Therefore, how to detect these qualities has been one of the researches focuses in food science.At present, many conventional methods based on sensory, chemical, physical, and microbiological approaches have been employed for detecting the quality changes of meat product. However, these methods always suffer certain disadvantages as they are normally ineffective, destructive, and/or high-cost, hindering them from further on-line applications. Consequently, it is of great significance to develop rapid, accurate, and non-destructive detection methods to identify the qualities of meat product. In order to implement this requirement, multispectral images of sausages treated by different processing methods and storage time were captured using multispectral imaging (MSI) system in the range of 405-970 nm, and the information among the acquired images were extracted and analyzed by chemometrics methods such as partial least square regression (PLSR), support vector machine (SVM), principal component analysis (PCA), successive projections algorithm (SPA), and gray level co-occurrence matrix (GLCM). The main achievements of this work are summarized as follow:(1) Based on spectral data, optimized models for hardness and aerobic plate count in sausages could be established by using SVM and PLSR methods, respectively. Thus, the good results of models were obtained with determination coefficient (Rp2) of 0.546 and ratio of prediction to deviation (RPD) of 1.455 for hardness, and Rp2 of 0.891 and RPD of 2.970 for aerobic plate count.(2) Combined textures of principal component images with spectra could improve precision of prediction models for moisture content, water holding capacity, cohesiveness, chewiness, heme rion content, and non-heme iron content in sausages. The Rp2 of these optimization models were 0.899,0.691,0.619,0.728,0.912, and 0.901, respectively, and the RPD of those were 3.027,1.800,1.561,1.770,3.356, and 3.167, respectively. Additionally, there was no correlation between the precision of prediction models and contribution rate of principal component.(3) Distribution of moisture content, water holding capacity, chewiness, aerobic plate count, heme iron content, and non-heme iron content in sausages could be successfully visualized by image processing and interpolation based on model equations. Thereby, the changes of sausage qualities could be visually evaluated by these maps, which demonstrated the superiority of MSI technology.This work suggested that MSI technique could apply to detect sausage qualities as a rapid, precious, and nondestructive method, which provided a feasibility approach for real-time detection of quality changes of sausages during processing and storage. Our results could provide a theory reference for meat industry application of MSI technology, and also reflect the important research value of that.更多还原