近红外光谱技术快速无损评价罗非鱼片新鲜度

Non-Destructive Freshness Evaluation of Tilapia(Oreochromis) Fillets Using Near Infrared Spectroscopy

利用傅里叶变换近红外光谱仪采集绞碎前后罗非鱼片背肉及腹肉的近红外光谱,并将其与总挥发性盐基氮(total volatile basic nitrogen,TVB-N)含量进行拟合,构建定量预测模型。在建模过程中,比较三点平滑、九点平滑(smoothing average 9 points,sa9)、九点卷积平滑(smoothing savitzky-golay 9 points,sg9)、一阶导数(1stderivative,Db1)、趋近归一化、单位长度归一化、标准正态变换、多元散射校正以及它们与Db1结合对光谱进行预处理的模型效果。结果表明,sg9和Db1相比于其他预处理方法可以较好地消除光谱噪音,提高模型预测能力,且各方法在与Db1联合使用后,模型的预测准确性以及建模效率普遍得到了提升。继续对光谱的波数范围进行筛选,剔除无关信息后,模型效果得到进一步提升,绞碎前背肉模型的校正集和验证集决定系数由0.870、0.821上升到了0.973、0.925,校正集和验证集标准偏差由2.152、2.991 mg/100 g减小到了1.032、1.581 mg/100 g。比较各模型效果可知,利用绞碎后的鱼肉光谱进行建模时效果要好于绞碎前的鱼肉。其中,以绞碎后腹肉模型的效果为最优,其验证集决定系数以及标准偏差分别为0.984、0.879 mg/100 g。但在综合考虑实际应用中快速、无损等需求后,绞碎前的鱼肉所建模型仍具有明显优势。最终,本研究选用绞碎前腹肉建立模型,校正集与验证集决定系数分别为0.982、0.976,校正集与验证集标准偏差分别为0.962、1.006 mg/100 g,在预测罗非鱼片TVB-N含量,快速、无损评价其新鲜度方面显示出了巨大潜力。

Fourier transform near infrared spectrometer was used in this experiment to collect the spectra of tilapia dorsal and belly muscle before and after being minced. By fitting the total volatile basic nitrogen(TVB-N) content to the spectra, quantitative prediction models were established. For modeling, Smoothing Average 3 Points(sa3), Smoothing Average 9 Points(sa9), Smoothing Savitzky-Golay 9 Points(sg9), 1st Derivative(Db1), Normalization by Closure(Ncl), Normalization to Unit Length(Nle), Standard Normal Variate(SNV), and Multiplicative Scatter Correction(MSC) were applied to pretreat the spectra. According to the results, sg9 and Db1 compared with other pretreatment methods could remove the noise, improve the prediction ability of models and the models showed better prediction accuracy and modeling efficiency by using other methods combined with Db1. The best wavenumber region was chosen to get rid of the irrelevant information and the models were further optimized. The determination coefficient of calibration set and validation set for dorsal muscle before being minced was increased from 0.870 and 0.821 to 0.973 and 0.925, respectively. While the standard errors were reduced to 1.032 and 1.581 mg/100 g from 2.152 and 2.991 mg/100 g, respectively. By comparison of model performance, the process of mincing was beneficial to modeling. And the model of minced belly muscle showed the best performance,which showed a determination coefficient of 0.984 with a standard error of 0.879 mg/100 g for validation set. But when the actual requirements for rapid and non-destructive freshness evaluation are under consideration, the model established for flesh before being minced still has obvious advantages. At last, the belly muscle before being minced was used to establishthe model. The calibration set gave a determination coefficient of 0.982 with a standard error of 0.962 mg/100 g and the validation set presented a determination coefficient of 0.976 with a standard error of 1.006 mg/100 g. This method showed enormous potential for TVB-N content prediction and non-destructive freshness evaluation of tilapia fillets.