摘要： 高光谱成像可同时获取被检测对象的图像信息和光谱信息，并对其内部成分进行定性和定量分析。国内外学者采用高光谱对肉品品质的研究多集中在水分、菌落总数、色泽、pH、挥发性盐基氮等方面，在肉品嫩度检测中应用区间变量迭代空间收缩法优选特征波长的研究鲜有报道。利用可见-近红外（400~1 000 nm）和近红外（900~1 700 nm）高光谱结合化学计量学方法对冷鲜滩羊肉嫩度进行无损预测，优选最佳建模波段。首先，采集羊肉的高光谱图像，提取样本感兴趣区域的光谱反射值，采用TA-XTplus质构仪测量滩羊肉嫩度；其次，将两个波段下的原始光谱数据进行多元散射校正（multiple scattering correction，MSC）、去趋势（de-trending）、基线校准（baseline）、标准正态变量（standard normal variable，SNV）、归一化（normalize）和卷积平滑（Savitzky-Golay）等预处理；分别采用连续投影算法（successive projection algorithm，SPA）、竞争性自适应加权算法（competitive adaptive reweighted sampling，CARS）、变量组合集群分析法（variables combination population analysis，VCPA）和区间变量迭代空间收缩法（interval variable iterative space shrinkage approach，IVISSA）对最佳预处理的光谱数据优选特征波长；最后，建立冷鲜滩羊肉嫩度的偏最小二乘回归（partial least squares regression，PLSR）预测模型，优选最佳建模波段。结果表明：（1）滩羊肉嫩度的近红外高光谱模型的预测效果优于可见-近红外高光谱；（2）经过多种预处理方法所建立的滩羊肉嫩度的模型中，近红外区域的原始光谱（original spectra，OS）模型效果最优，其Rc=0.83，Rp=0.79，RMSEC=874.94，RMSEP=1 465.97；（3）近红外高光谱的原始光谱经SPA，CARS，VCPA，IVISSA四种方法共挑选出15，16，13和123个特征波长，占总波长的7%，6%，5%和54% ；（4）近红外高光谱结合OS-IVISSA-PLSR建立的冷鲜滩羊肉嫩度预测模型最好，其Rc=0.85, RMSEC=850.86，Rp=0.79，RMSEP=1 497.11。IVISSA算法不仅可大幅度减少模型运算次数，还可以保证模型的精准和稳定性。研究表明，OS-IVISSA-PLSR模型对冷鲜滩羊肉嫩度进行高光谱的快速无损检测是可行的。
关键词：冷鲜滩羊肉；嫩度；高光谱成像技术；区间变量迭代空间收缩法；偏最小二乘回归
Abstract：Hyperspectral imaging can obtain both the image information and spectral information of the detected object, and make qualitative and quantitative analysis with internal components. The research on meat quality by hyperspectral imaging technology focuses on water, total viable count, color, pH, total volatile basic nitrogen etc. There are few studies on meat tenderness based on interval variable iterative space contraction method. In this paper, tenderness values of chilled Tan-sheep were detected non-destructively by visible-near-infrared and near-infrared bands (400~1 000, 900~1 700 nm) combined with chemometric methods to obtain the best modeling bands. Firstly, hyperspectral images of lamb sample were collected and extracted the spectral values of the region of interest; lamb tenderness was measured using a TA-XTplus texture analyzer; Secondly, the original spectral data between 400～1 000 and 900～1 700 nm were preprocessed by multiple scattering correction (MSC), de-trending, standard normal variate (SNV), baseline, normalize, Savitzky-Golay; optimal bands were selected by successive projection algorithm (SPA), competitive adaptive reweighted sampling (CARS), variables combination population analysis (VCPA) and interval variable iterative space shrinkage approach (IVISSA). Finally, optimal bands were selected and partial least squares regression (PLSR) model of chilled mutton was established. The results showed that: (1) the prediction performance of original spectral model in the region of 900～1 700 nm was better than that of 400～1 000 nm. (2) original spectral mode of the tenderness of chilled Tan-sheep had the best performance Rc=0.83, Rp=0.79, RMSEC=874.94, RMSEP=1 465.97 in the near-infrared region using different pretreatment methods. (3) the original spectrum of 900～1 700 nm was selected by SPA, CARS, VCPA and IVISSA with 15, 16, 13 and 123 characteristic wavelengths, accounting for 7%, 6%, 5% and 54% of the total wavelength. (4) the prediction model of chilled Tan-sheep tenderness was the best in combination with hyperspectral technique and OS-IVISSA-PLSR with Rc=0.85, RMSEC=850.86, RMSECV=1 193.42, Rp=0.79, RMSEP=1 497.11. It was indicated that IVISSA algorithm could greatly reduce the number of model operations and ensure the predictability and stability of the model. It is feasible to adopt OS-IVISSA-PLSR method to analyze the tenderness of chilled Tan-sheep mutton.
Key words：Chilled Tan-sheep mutton; Tenderness; Hyperspectral imaging technology; Interval variable iterative space shrinkage approach; Partial least squares regression
刘贵珊，张 翀，樊奈昀，程丽娟，余江泳，袁瑞瑞. IVISSA算法冷鲜滩羊肉嫩度的高光谱模型优化[J]. 光谱学与光谱分析, 2020, 40(08): 2558-2563.
LIU Gui-shan, ZHANG Chong, FAN Nai-yun, CHENG Li-juan, YU Jiang-yong, YUAN Rui-rui. Hyperspectral Model Optimization for Tenderness of Chilled Tan-Sheep Mutton Based on IVISSA. SPECTROSCOPY AND SPECTRAL ANALYSIS, 2020, 40(08): 2558-2563.
[1] WANG Song-lei, WU Long-guo, MA Tian-li, et al(王松磊, 吴龙国, 马天利，等). Spectroscopy and Spectral Analysis(光谱学与光谱分析), 2016, 36(9): 2937.
[2] SONG Hong-xia, HUANG Feng, DING Zhen-jiang, et al(孙红霞, 黄 峰, 丁振江, 等). Food Science(食品科学), 2018, 39(1): 84.
[3] Sun H, Peng Y, Zheng X, et al. Journal of Food Engineering, 2019, 248: 1.
[4] ZHAO Juan, PENG Yan-kun(赵 娟, 彭彦昆). Journal of Agricultural Engineering(农业工程学报), 2015，31(7)：279.
[5] Lee Hoonsoo, Kim M S, Lee W H, et al. Sensors and Actuators B: Chemical, 2018, 259: 532.
[6] Achata E M, Inguglia E S, Esquerre C A, et al. Journal of Food Engineering, 2019, 246: 134.
[7] Qu F, Ren D, He Y, et al. Meat Science, 2018, 146: 59.
[8] Lee H, Kim M S, Lee W, et al. Sensors and Actuators B: Chemical, 2018, 259: 532.
[9] Zhang D, Xu Y, Huang W, et al. Infrared Physics & Technology, 2019, 98: 297.
[10] CHENG Li-juan, LIU Gui-shan, HE Jian-guo, et al(程丽娟, 刘贵珊, 何建国，等). Food Science(食品科学)，2019, 40(10): 285.
[11] ZHAO Huan, HUAN Ke-wei, SHI Xiao-guang, et al(赵 环, 宦克为, 石晓光，等). Analytical Chemistry(分析化学)，2018, 46(1): 136.
[12] Song X, Huang Y, Yan H, et al. Analytica Chimica Acta, 2016, 948: 19.
[13] WU Long-guo, HE Jian-guo, LIU Gui-shan, et al(吴龙国, 何建国, 刘贵珊, 等). Photoelectron Laser(光电子激光), 2014，25(1): 135.
[14] Ma J, Sun D, Pu H, et al. Journal of Food Engineering, 2019, 240: 207.
[15] WANG Wan-jiao, WANG Song-lei, HE Xiao-guang, et al(王婉娇, 王松磊, 贺晓光，等). Science and Technology of Food Industry(食品工业科技), 2015, 36(20): 77.
