熟驴肉掺假近红外定性及定量检测

Qualitative and Quantitative Analyses of Cooked Donkey Meat Adulteration Based on NIR Spectroscopy

驴肉风味绝佳、营养丰富,价格高昂且供应量低。驴肉火烧等熟驴肉中掺假马肉、骡子肉等其他肉类的问题亟待解决。为实现熟驴肉样品在不同掺假比例下的定性定量分析,以10%为梯度,分别制备驴肉含量为0%~100%的驴马掺假和驴骡掺假样品,并在4000~12500 cm^(-1)光谱范围采集样品光谱。针对熟驴肉掺假问题,采用线性判别分析、支持向量机以及广义回归神经网络的方法结合平滑算法(5点、15点、25点)、多元散射校正(MSC)、标准正态变量变换(SNV)、基线校正(Baseline)、归一化和去趋势化(Detrend)等预处理方法建立了近红外鉴别模型。采用偏最小二乘回归(PLSR)、反向传播神经网络(BP)方法结合以上预处理方法分别对熟驴肉掺假样品建立定量模型检测驴肉含量。SNV预处理结合支持向量机对熟肉碎掺假样品鉴别模型结果较优,校正集和预测集判别正确率为98.70%和94.78%;Detrend预处理结合线性判别分析的熟肉糜掺假样品鉴别模型结果较优,校正集和预测集的判别正确率分别达到98.47%和96.23%。建立定量模型进行掺假样品中驴肉含量的检测,BP模型相较于PLSR模型均取得较好的结果,具有较高的决定系数(R^(2))、相对分析误差(RPD)和较低的均方根误差(RMSE)。对于熟肉碎掺假样品,驴骡掺假样品集在Detrend处理后的BP模型结果较好,交叉验证集以及预测集的R^(2)、RMSE、RPD分别为0.971、0.067、5.844,0.980、0.086、6.984;驴马掺假样品集在归一化预处理后BP模型结果较好,各项参数分别为0.997、0.032、18.026,0.982、0.089、7.454。对于熟肉糜掺假样品,经Detrend预处理的BP模型结果均较优,驴骡掺假样品最佳定量模型的参数分别为0.982、0.041、7.470,0.986、0.103、8.452;驴马掺假最佳模型参数分别为0.986、0.036、8.348,0.961、0.101、5.044。结果表明,近红外光谱结合不同的建模算法实现了不同比例掺假的熟驴肉样品快速无损检测和鉴别,能够用于此后熟驴肉掺假的定性、定量分析。

Donkey meat has excellent flavor and rich nutrition and is in high price and low supply.The problem of cooked donkey meat adulterated with other meat,such as horse and mule meat,needs to be solved urgently.To realize the qualitative and quantitative analysis of cooked donkey meat samples of different adulteration ratios,horse and mule meat samples were used to degrade donkey meat.The gradient was 10%,and the donkey meat contents were 0%~100%.Spectra of samples were collected in the range of 4000~12500 cm^(-1).The methods of linear discriminant analysis,support vector machine,and generalized regression neural network combined with smoothing algorithm(5 points,15 points,25 points),multiplicative scattering correction(MSC),standard normal variable(SNV),Baseline correction,normalization,and Detrend were used to establish the NIR discriminant models of adulterated cooked donkey meat samples.Partial least squares regression(PLSR)and backpropagation(BP)were used to establish quantitative models to determine the content of donkey meat in adulterated samples.For minced after cooked meat samples,the results of SNV pretreatment combined with a support vector machine were optimal,and the discriminant accuracy of the calibration set and prediction set was 98.70% and 94.78%.The results of Detrend pretreatment combined with linear discriminant analysis were optimal for minced before cooked meat samples.The discriminant accuracy of the calibration and prediction sets reached 98.47% and 96.23%,respectively.Compared with the PLSR model,the BP model obtained better results,with a higher coefficient of determination(R^(2)),relative percent deviation(RPD),and lower root mean square error(RMSE).For the adulterated samples of minced after cooked meat samples,the BP model of the donkey and mule adulterated samples was better after Detrend pretreatment.R^(2),RMSE,and RPD of the cross-validation set and prediction set were 0.971,0.067,5.844,0.980,0.086,6.984,respectively.After normalized treatment,the results of BP model of donkey and horse adulterated samples were optimal,and the parameters were 0.997,0.032,18.026,0.982,0.089,7.454,respectively.For the adulterated samples of minced before cooked meat samples,the results of the BP model with Detrend pretreatment were better,and the optimal quantitative model parameters of donkey and mule adulterated samples were 0.982,0.041,7.470,0.986,0.103,8.452,respectively.The best model parameters of donkey and horse adulteration were 0.986,0.036,8.348,0.961,0.101,and 5.044,respectively.The results show that the NIR spectroscopy combined with different modeling algorithms can realize the rapid,nondestructive detection of different donkey meat contents.The methodology can be used for future qualitative and quantitative analysis of cooked donkey meat adulteration.