近红外高光谱联用Stepwise算法快速无接触评估冷鲜鸡肉色泽及嫩度

利用900~1700nm近红外高光谱成像系统联用Stepwise算法快速评估鸡肉色泽和嫩度。通过采集新鲜屠宰鸡肉高光谱图像,提取试验样本感兴趣区域(Region of interests,ROI)反射光谱信息,经中值滤波平滑(Median filtering smoothing,MFS)、多元散射校正(Multiplicative scatter correction,MSC)和标准正态变量变换( Standard normal variable correction,SNV)三种预处理后,分别利用偏最小二乘(Partial Least Squares,PLS)和多元线性回归(Multiple linear regression,MLR)挖掘光谱信息与鸡肉色泽参数(L^*、a^*、b^*)及嫩度参考值之间的定量关系。结果显示,经MFS预处理的近红外光谱(486个波长)构建的全波段PLS回归模型(F-PLS)预测L^*(RP=0.904,RMSEP=2.036)、b^*(RP=0.908,RMSEP=1.577)和嫩度(RP=0.948,RMSEP=1.596)效果更好。为提高预测效率,采用Stepwise算法筛选最优波长优化F-PLS模型,结果显示,从SNV预处理光谱筛选的14个最优波长构建MLR回归模型预测L^*值(RP=0.894,RMSEP=2.160)效果较优,从SNV预处理光谱筛选的13最优波长构建的O-PLS回归模型预测b^*值(RP=0.877,RMSEP=1.811)效果较优,从MFS预处理光谱筛选的20个最优波长构建O-PLS回归模型预测嫩度值(RP=0.888,RMSEP=2.408N)效果较优。本试验表明,利用近红外高光谱成像技术结合Stepwise算法可实现鸡肉色泽参数L^*、b^*值以及嫩度的快速评估。

Rapid and Non-Contact Evaluation of Color and Tenderness in Fresh Chilled Chicken by Near-Infrared Hyperspectral Imaging Combined with Stepwise Algorithm

The color and tenderness of fresh chilled chicken were evaluated by near-infrared hyperspectral imaging system( 900 ~1700 nm) combined with stepwise algorithm.By collecting hyperspectral images of fresh slaughtered chicken,extracting spectral reflectance information within the region of interests( ROI) of images of test samples,pretreating spectra with median filter( MFS),multivariate scattering correction ( MSC) and standard normal variable transformation ( SNV),respectively,the spectral data was mined by partial least square ( PLS) and multivariate linear regression ( MLR) to build the quantitative relationship between spectra and chicken color parameters( L^*,a^*,b^*) and tenderness.As a result,the PLS regression based on full 486 wavelengths ( F- PLS) pretreated by MFS showed better performance in predicting L^*( RP = 0.904,RMSEP = 2.036),b^*( RP = 0.908,RMSEP = 1.577) and tenderness( RP = 0.948,RMSEP = 1.596).The optimal wavelengths were then selected by stepwise algorithm to simplify the F-PLS models and improve the prediction efficiency.It was indicated that MLR model established with 14 optimal wavelengths selected from SNV spectra showed better performance in predicting L^* value ( RP = 0.894,RMSEP = 2.160).The O-PLS model based on 13 optimal wavelengths screened from SNV pretreatment spectrum had better performance in predicting b^* value ( RP = 0.877,RMSEP = 1.811).The O- PLS regression model based on 20 optimum wavelengths screened from MFS pretreatment spectrum showed better performance for predicting tenderness( RP = 0.888,RMSEP = 2.408 N).The overall results indicated that near - infrared hyperspectral imaging combined with stepwise algorithm had a great potential in fast evaluation of color( L^* and b^* value) and tenderness in chicken.