针对明虾品质劣变过程中的鲜度特征, 该研究以颜色(L^*, a^*, b^*)、 挥发性盐基氮(TVB-N)、 pH等品质指标为研究对象, 利用拉曼无损检测技术采集4 ℃和-20 ℃下生鲜明虾光谱信息, 运用岭回归、 偏最小二乘法和前向逐步回归对明虾进行了快速定量检测, 建立了品质指标定量模型。 其中光谱数据预处理包括SG平滑、 背景扣除、 二阶微分、 标准正态变量变换, 按一定方式组合4种预处理和PCA降维技术进行数据处理, 筛选出最优模型。 结果表明: 利用岭回归建立颜色(a^*, b^*)定量模型时, 在组合预处理方式下建模集中 R分别为0.983和0.973, RMSE分别为0.114和0.179, 预测集中 R分别为0.513和0.564; RMSE分别为0.615和0.918, 建模集精度远超预测集表明出现过拟合, 经PCA降维后过拟合降低、 但预测集预测效果不理想; 偏最小二乘法在各指标建模集上精度和岭回归差不多, 在预测集上预测精度偏低, PCA降维后部分指标建模集相关系数下降、 均方根误差上升, 预测精度降低。 最终结果显示: 经过4种预处理后的前向逐步回归模型最优, 建模集中L^*, a^*, b^*, pH和TVB-N指标 R分别为0.904, 0.885, 0.864, 0.934和0.940, RMSE分别为1.141, 0.280, 0.535, 0.131和2.345; 预测集中 R分别为0.863, 0.850, 0.859, 0.900和0.916, RMSE分别为1.394, 0.406, 0.605, 0.194和2.734, 建模效果好。 因此, 利用拉曼光谱技术结合前向逐步回归模型快速检测明虾中L^*, a^*, b^*, pH和挥发性盐基氮含量可行, 对拉曼技术应用明虾品质检测具有一定指导意义。

About the prawn’s freshness characteristics of quality deterioration, the research takes color (L^*, a^*, b^*), volatile base nitrogen(TVB-N), ph, and other quality indexes as the object of the study, and uses Raman nondestructive testing technology to select the spectral information of fresh prawn on the temperature of 4 ℃ and under -20 ℃, also makes the quick quantitative test by combining with ridge regression, partial least squares method and forward stepwise regression, establishes the quantitative mode of the quality index. And the spectral data preprocessing includes SG smoothing, background deduction, second order differential and standard normal variable transform, combines 4 types of preprocessing in a certain way and deals with the data by PCA dimension reduction technology, in order to select the best mode. The result shows that, when using ridge regression to establish the quantitative mode of color (a^*, b^*), under the combined pretreatment mode, the modeling centralization R are 0.983 and 0.973 respectively, RMSE are 0.114 and 0.179 respectively; the forecast concentration R are 0.513 and 0.564 respectively, RMSE are 0.615 and 0.918 respectively, the accuracy of the modeling set is much higher than that of the prediction set, which indicates that there exists over-fitting, and the over-fitting decreases after dimension reduction by PCA, but the prediction effect of prediction sets is not satisfactory; partial least squares method and the ridge regression are about the same on the accuracy of indicator modeling sets, the accuracy of partial least squares method is lower on the prediction sets. After PCA dimension reduction, the related coefficient of partial index modeling sets decrease, the root mean square error increases, and the prediction accuracy decreases. The final result shows that, after 4 types of preprocessing, the mode of forward stepwise regression is the best, the modeling centralization L^*, a^*, b^*, pH, TVB-N index R are 0.904, 0.885, 0.864, 0.934, 0.940 respectively, RMSE are 1.141, 0.280, 0.535, 0.131, 2.345 respectively; the forecast concentration R are 0.863, 0.850, 0.859, 0.900, 0.916 respectively, RMSE are 1.394, 0.406, 0.605, 0.194, 2.734 respectively, the modeling effect is good. Therefore, it is practicable to use the Raman spectroscopy technology, combining with forwarding stepwise regression to quick test the prawn’s L^*, a^*, b^*, pH and volatile base nitrogen content, which provide meaningful guidance for the application of Raman technology in prawn quality detection.