Rapid and Non-Destructive Detection of Tan Sheep Meat MetMb Contents Using Hyperspectral Imaging
CHENG Li-juan1, LIU Gui-shan1*, HE Jian-guo1, WAN Guo-ling1, MA Chao2, BAN Jing-jing1, MA Li-min1, YANG Guo-hua1, YUAN Rui-rui1
1. School of Agriculture Department of Food, Ningxia University, Yinchuan 750021, China
2. School of Physics and Electrical and Electronic Engineering, Ningxia University, Yinchuan 750021, China
摘要： 高铁肌红蛋白（metmyoglobin, MetMb）在肉中所占的比例直接影响肉的色泽。利用可见近红外光谱（ViS-NIR）采集到的滩羊肉数据与化学计量学方法相结合，探讨高光谱成像快速无损检测滩羊肉中MetMb含量的可行性以及开发滩羊肉中MetMb含量的定量函数。采用分光光度计测量滩羊样本的MetMb含量，使用ENVI4.8软件提取贮藏期间200个样本光谱图像的感兴趣区域，将获取的光谱数据与化学值相结合，定量解释两者的相关性；利用光谱理化值共生距离法，按照3∶1的比例划分样本，对校准模型进行独立（外部）验证；采用乘法散射校正（multiple scattering correction，MSC）、一阶导数 (first derivative, 1st derivative)和去趋势(De-trending)等3种不同的方法对原始光谱数据进行预处理，以消除噪音对原始光谱的干扰；竞争性自适应加权算法(competitive adaptive reweighted sampling，CARS)、区间变量迭代空间收缩方法（interval variable iterative space shrinkage approach, iVISSA)、间隔随机蛙跳算法（interval random frog，IRF）、变量组合集群分析法(variables combination population analysis，VCPA)、连续投影算法(successie projection algorithm，SPA)以及IRF+SPA、iVISSA+SPA组合方法被用于光谱的变量选择和优化；使用典型的线性建模方法：偏最小二乘回归(partial least square regression, PLSR)建立全波段和特征波段的预测模型，确定最佳模型；通过最佳模型建立滩羊肉中MetMb含量的定量函数。结果表明：原始光谱模型性能较好于3种预处理光谱的模型性能，其R2C=0.852，R2P=0.788，RMSEC=4.604，RMSEP=5.729；原始光谱经过CARS，VCPA，IRF，SPA，iVISSA，IRF+SPA，iVISSA+SPA等方法分别选出16，13，48，14，45，10和11个特征波长，占总波长的12.8%，10.4%，38.4%，11.2%，36%，8%和8.8%。通过对比PLSR模型，IRF+SPA-PLSR模型性能最佳，R2C=0.808，R2P=0.826，RMSEC=5.253，RMSEP=5.149，IRF+SPA算法不仅减少了计算时间，而且生成了更准确，更稳健的预测模型；最后，基于IRF+SPA算法建立的MetMb含量的定量函数为：(MetMb)=3.249 7+1.900 2λ468-4.791 2λ482+5.913 5λ512-1.856 2λ530-5.879 3λ545+2.246 3λ560+5.066 1λ580-2.320 1λ588+1.214 9λ790-1.348 8λ814。表明ViS-NIR光谱对滩羊肉中MetMb含量的快速无损检测是可行的，开发的定量函数为快速测定滩羊肉中MetMb的含量提供参考。
关键词：滩羊；高铁肌红蛋白；可见/近红外；偏最小二乘回归；特征波长选择
Abstract：The proportion of Metmyoglobin (MetMb) in meat directly affects the color of the meat. This paper combined the visible near-infrared spectroscopy (ViS-NIR) data of Tan sheep meat with the chemometric method to explore the feasibility of rapid non-destructive detection of MetMb content in Tan sheep by hyperspectral imaging technology and develop a quantitative function of MetMb content. The MetMb content of the sample was measured by a spectrophotometer, and the interest region of 200 sample spectral images during storage were extracted by ENVI4.8 software. The relationships between MetMb content and spectral date of samples were established to quantitatively analyze. In this study, according to the ratio of 3∶1, the whole dataset (n=200) was divided into a calibration set (n=50) for performing independent validation (external validation) of the developed calibration models using the sample set partitioning based on joint x-y distance method. Several spectral preprocessing techniques such as multiplicative scatter correction (MSC), first derivative (1st derivative) and De-trending were applied to eliminate noise. Competitive Adaptive Reweighted Sampling (CARS), Interval variable iterative space shrinkage approach (iVISSA), Interval Random Frog (IRF), Variables combination population analysis (VCPA) and Successie Projection Algorithm (SPA) were used to select and optimize variables. Partial least squares regression (PLSR), which was one classical linear calibration method, were used for developing prediction models based on full-band and feature bands. The results showed that the original spectral model was best, and its R2C=0.852, R2P=0.788, RMSEC=4.604, RMSEP=5.729. The CARS, IRF, SPA, iVISSA, VCPA, IRF+SPA and iVISSA+SPA methods were applied to select 16, 13, 48, 14, 45, 13, 10 and 11 feature wavelengths from the original spectra, accounting for 38.4%, 10.4%, 11.2%, 36%, 10.4%, 8%, 8.8% and 12.8% of the full wavelength, respectively. The IRF+SPA-PLSR model was the best among the models developed, and its R2C, R2P, RMSEC and RMSEP values were 0.808, 0.826, 5.253 and 5.149, respectively. The IRF+SPA algorithm greatly reduced calculating time and generated more accurate and more robust prediction model compared with full band. Finally, the quantitative linear relationship between spectral data and MetMb parameters was established based on the IRF+SPA algorithm, and the quantitative function was: (MetMb)=3.249 7+1.9002λ468-4.791 2λ482+5.913 5λ512-1.856 2λ530-5.879 3λ545+2.246 3λ560+5.066 1λ580-2.320 1λ588+1.214 9λ790-1.348 8λ814. It is shows that Vis-NIR is feasible for the rapid non-destructive detection of MetMb content in Tan sheep. Simultaneously, the quantitative function developed provides a reference for the rapid determination of MetMb content in Tan sheep.
程丽娟，刘贵珊，何建国，万国玲，马 超，班晶晶，马丽敏，杨国华，袁瑞瑞. 基于高光谱开发滩羊肉中高铁肌红蛋白含量的定量函数[J]. 光谱学与光谱分析, 2020, 40(04): 1263-1269.
CHENG Li-juan, LIU Gui-shan, HE Jian-guo, WAN Guo-ling, MA Chao, BAN Jing-jing, MA Li-min, YANG Guo-hua, YUAN Rui-rui. Rapid and Non-Destructive Detection of Tan Sheep Meat MetMb Contents Using Hyperspectral Imaging. SPECTROSCOPY AND SPECTRAL ANALYSIS, 2020, 40(04): 1263-1269.
[1] Cammarata M B, Brodbelt J S. Chem. Sci.，2015，6：1324.
[2] Wright T J, Davis R W. Protein Expression and Purification, 2015, 107: 50.
[3] Taghdisi S M, Danesh N M, Ramezani M, et al. Biosensors and Bioelectronics, 2016, 80: 532.
[4] Osman B, Uzun L, Besirli N, et al. Mater. Sci. Eng. C，2013，33：3609.
[5] Velásquez L, Cruz-Tirado J P, Siche Raúl, et al. Meat Science, 2017, 133: 43.
[6] Kamruzzaman M, Makino Y, Oshita S. Meat Science, 2016, 116: 110.
[7] Qiao L, Tang X, Dong J. Food Chemistry, 2017, 237: 1179.
[8] Ma J, Pu H, Sun D W. LWT, 2018, 94: 119.
[9] Yang D, He D, Lu A, et al. Infrared Physics & Technology, 2017, 83: 206.
[10] Nubiato K E Z, Mazon M R, Antonelo D S, et al. Infrared Physics & Technology, 2018, 89: 247.
[11] Kamruzzaman M, Makino Y, Oshita S. Meat Science, 2016, 116: 110.
[12] Kamruzzaman M, Makino Y, Oshita S. Journal of Food Engineering, 2016, 170(7): 8.
[13] YANG Xiao-yu, LIU Gui-shan, DING Jia-xing, et al（杨晓玉, 刘贵珊, 丁佳兴, 等）. Spectroscopy and Spectral Analysis(光谱学与光谱分析), 2019, 39(1): 230.
[14] Cozzolino D, Murray I. LWT-Food Science and Technology, 2004, 37(4): 447.
[15] Mamani-Linares L W, Gallo C, Alomar D. Meat Science, 2011, 90(2): 378.
[16] Cheng J H, Sun D W. LWT-Food Science and Technology, 2015, 62(2): 1060.
