牛肉质构特性的近红外光谱无损检测

Nondestructive determination of beef textural properties by near infrared spectroscopy

为了建立基于近红外光谱技术的牛肉质构特性快速检测方法,该试验采集了202个新鲜牛肉样品在800~2 500 nm波长范围内的漫反射光谱,测定了牛肉的硬度、弹性、咀嚼性和黏附性,经小波消噪后,分别采用平滑、一阶微分、二阶微分等6种方法预处理,建立了牛肉质构特性的偏最小二乘回归模型,并用最优模型进行预测。结果表明：经小波消噪后采用二阶微分预处理方法建立的牛肉硬度、弹性、咀嚼性的检测模型效果最好,其校正集相关系数r均在0.9以上,校正集均方根误差（root means square error of calibration,RMSEC）分别为6.247 N、0.760 mm、14.954 m J,预测集相关系数均在0.664以上,预测集均方根误差（root means square error of prediction,RMSEP）分别为8.887 N、0.951 mm、22.117 m J,相对预测误差（ratio of prediction to deviation,RPD）值分别为2.43、1.88、2.32,预测精度较高,能够有效地预测牛肉的硬度、咀嚼性,可以检测精度要求不高的牛肉弹性;试验所建立的牛肉黏附性检测模型的预测性能不是很理想,虽然其校正集和预测集相关系数较高（分别为0.720、0.694）,RMSEC和RESEP均较小（分别为0.302、0.243 N·mm）,但其RPD值小于1.5,模型预测精度较差,不可以用于预测未知样品的黏附性,此方法还需进一步研究。研究结果为牛肉质构特性的快速无损评价提供了理论依据。

This study aims to establish a rapid detective method for the characteristics of beef texture through near infrare spectroscopy. The diffuse reflectance spectra at 800-2500 nm of 202 fresh beef samples were collected. The hardness springiness, chewiness and the adhesiveness of beef samples were tested by texture analyzer. After wavelet denoising, the dat were collected through processing the diffuse reflectance spectra by smoothness, first order differential, second orde differential, standard normal variate（SNV）, SNV combined with first order differential or SNV combined with second orde differential respectively, and then the partial least square regression（PLSR） statistical analysis models for the near-infrare spectra of beef samples were established. According to the predicted results of hardness, springiness and chewiness of bee through 6 PLSR statistical analysis models, the optimal predictive model was the PLSR statistical analysis model for th near-infrared spectra by the second order differential preconditioning method after wavelet denoising. The correlatio coefficients of their calibration sets for hardness, springiness and chewiness of beef samples were all above 0.9, and the roo mean square errors of calibration（RMSEC） were 6.247 N, 0.760 mm and 14.954 mJ, respectively. The correlation coefficient of their prediction sets were all above 0.664, and the root mean square errors of prediction（RMSEP） were 8.887 N, 0.951 mm and 22.117 mJ, respectively. The ratio of prediction to deviation（RPD） were 2.43, 1.88 and 2.32, respectively. The hig predictive accuracy of hardness and chewiness of beef samples was obtained through the established optimal model, whic was also fit for predicting the springiness of beef with low accuracy. The hardness, springiness and chewiness of beef ar closely related to its own water, protein, fat and other chemical components. These chemical components of beef can b detected using near infrared spectra. Therefore, the hardness, springiness and chewiness of beef can be predicted through nea infrared spectroscopy technology. The correlation coefficients of the adhesiveness in the calibration set and prediction set wer high（0.720 and 0.694）, and RMSEC and RESEP were small（0.302 and 0.243 N·mm） by using first order differentia preconditioning method after wavelet denoising, but its RPD value was less than 1.5, so the prediction performance of th detection model of beef adhesiveness was not very satisfactory and the model was not fit for predicting the adhesiveness o unknown samples. This may be due to the adhesiveness value of each sample was the average value of 4 points in differen parts of the sample. The difference was relatively large between the measured values, and the performance of the predictio model was affected by the accuracy of the measurement results. Further studies were needed for establishing the accurac predictive model of the adhesiveness of beef by near infrared spectroscopy. The present study has established a rapid an accurate predictive model, which possesses the features of nondestructive inspection, qualitative analysis and quantitativ detection for beef texture characteristics.