光谱学与光谱分析, 2019, 39 (11): 3365, 网络出版: 2019-12-02  

空间分辨散射光谱多参数信息融合方法的生鲜肉嫩度无损检测

Nondestructive Detection of Pork Tenderness Using Spatially Resolved Hyperspectral Imaging Technique Based on Multivariable Statistical Analysis
作者单位
中国农业大学工学院, 国家农产品加工技术装备研发分中心, 北京 100083
摘要
嫩度是猪肉食用品质最重要的指标之一。 猪肉嫩度取决于猪肉组织复杂的物理、 化学特性, 目前难以实现快速无损伤检测。 探索空间分辨光谱技术用于生鲜肉嫩度无损检测的可行性。 首先利用点光源高光谱扫描系统采集54块猪肉背最长肌的空间可分辨散射光谱, 经过感兴趣区域选择, 提取出猪肉样本表面光斑的空间扩散轮廓, 结合4-参数洛伦兹分布函数对扩散轮廓进行非线性拟合, 拟合优度R2>0.992, 并通过残差分析, 表明4-参数洛伦兹分布函数符合肉样表面光强的空间散射规律, 进而提取出480~950 nm波长下空间分辨光谱的四个形态学参数: 渐进值a、 峰值b、 半带宽c以及半带宽处的斜率d。 然后将单参数谱分别与猪肉样本Warner-Bratzler剪切力(WBSF)测量值进行偏最小二乘回归(PLSR)分析。 结果表明不同参数谱都含有猪肉嫩度信息, 其中峰值参数谱b建模效果最佳, 其回归模型的校正集决定系数R2c为0.674, 均方根误差SEC为8.396N, 预测集决定系数R2p为0.610, 均方根误差SEP为8.643N。 为提高模型的预测精度和稳定性, 实现多参数谱信息的融合, 先通过PLSR分析, 分别提取出每个参数谱中对猪肉嫩度方差贡献大的公共因子, 然后将其因子得分组合在一起作为参数谱的特征变量, 与猪肉样本WBSF测量值作多元统计回归分析。 为避免数据冗余, 对不同参数谱特征变量进行多重共线性判别, 进一步采用PLSR算法对参数谱特征变量进行降维和变换, 采用交叉验证方法, 选择前两维因子得分进行校正模型的建立。 其中所提取第一维公共因子对猪肉WBSF值方差解释率达92.28%。 与单参数谱所建PLSR模型相比, 多参数谱信息融合模型预测效果有了较大提高, 其R2c和R2p分别为0.923和0.800, SEC和SEP分别为4.083N和5.655N。 通过对回归系数进行统计量t检验, 结果表明所有回归系数极显著(p<0.01)。 本研究通过采取多参数信息融合方法为空间分辨光谱在生鲜肉嫩度无损检测应用提供一种思路, 该方法有效将空间分辨光谱解析为4个形态学参数, 并实现不同参数谱信息的提取和融合, 为开发基于空间分辨光谱的生鲜肉嫩度无损快速检测装备提供技术支撑。
Abstract
Tenderness is one of the most important attributes of pork eating quality. The tenderness of pork depends on the complex physical and chemical characteristics of pork tissue. And a rapid, non-destructive detection method is urgently in need. This paper reports the feasibility of spatially resolved hyperspectral imaging technique for nondestructive detection of pork tenderness. First, the spatial resolved scattering images of 54 pork longissimus dorsi muscle were collected by hyperspectral system on line-scanning mode. The region of interest (ROI) was selected and the diffused spatial profile of incident light was extracted on the surface of the pork sample. The diffused spatial profile was fitted non-linearly by 4-parameter Lorentzian distribution function. The goodness of fit was R2>0.992, and the residual analysis showed that the 4-parameter Lorentzian function could describe the spatial distribution of light intensity on meat surface. Four morphological parameters of spatial resolved spectrum at wavelength of 480~950 nm were extracted: asymptotic value a, peak value b, full width at half of the peak value c (FWHM) and slope at half of the peak value d. Partial least squares regression (PLSR) models were established to relate each parameter spectra and Warner Bratzler shear force (WBSF) values of pork samples respectively. The results showed that all parameters spectra contained pork tenderness information, in which the peak parameter b had the best prediction results, with determination coefficient of calibration set R2c of 0.674, the root-mean-square error SEC of 8.396N, the determination coefficient of prediction set R2p of 0.610, and the root-mean-square error SEP of 8.643N. In order to improve the accuracy and stability of the prediction model and realize the information fusion of multi-parameter spectra, PLSR analysis was firstly used to extract the latent variables in each parameter spectrum, which have high relative variance contributionto pork tenderness. Then, the latent variable scores were combined as the characteristic variables of the parameter spectra, and multiple statistical regression analysis was performed to relate the characteristic variables and the WBSF values of pork samples. In order to avoid data redundancy, PLSR algorithm was secondly used to reduce and transform the characteristic variables of the parameter spectra. Using the cross validation method, the first two - dimensional factor scores were selected to establish the calibration model. The variance interpretation rate of the pork WBSF value from the first factor was 92.28%. Compared with the PLSR model built by the single-parameter spectrum, the prediction results of the multi-parameter spectra model have been greatly improved, with R2c of 0.923 and R2p of 0.800, SEC of 4.083N and SEP of 5.655N respectively. The results show that all regression coefficients are very significant (p<0.01). In this study, the multi-parameter information fusion method was adopted to provide an idea for the application of spatial resolution spectroscopy in the nondestructive testing of pork tenderness. This method decomposed the spatial resolved spectra into 4 morphological parameters effectively, and achieved the information extraction and fusion of different parameter spectra, providing technical support for the development of non-destructive rapid detection equipment for pork tenderness based on spatial resolved spectroscopy technology.

孙宏伟, 彭彦昆, 王凡. 空间分辨散射光谱多参数信息融合方法的生鲜肉嫩度无损检测[J]. 光谱学与光谱分析, 2019, 39(11): 3365. SUN Hong-wei, PENG Yan-kun, WANG Fan. Nondestructive Detection of Pork Tenderness Using Spatially Resolved Hyperspectral Imaging Technique Based on Multivariable Statistical Analysis[J]. Spectroscopy and Spectral Analysis, 2019, 39(11): 3365.

关于本站 Cookie 的使用提示

中国光学期刊网使用基于 cookie 的技术来更好地为您提供各项服务,点击此处了解我们的隐私策略。 如您需继续使用本网站,请您授权我们使用本地 cookie 来保存部分信息。
全站搜索
您最值得信赖的光电行业旗舰网络服务平台!