AFB_(1)污染花生油激光诱导荧光信号受温度干扰规律及温度全局模型研究

Temperature Interference Law of Laser-Induced Fluorescence Signal of AFB_(1) Contaminated Peanut Oil and Global Temperature Model

激光诱导荧光光谱(LIF)技术在花生、无花果等坚果中黄曲霉毒素B_(1)(AFB_(1))检测应用已有大量研究。然而荧光信号易受环境因素(主要是温度)影响。本研究通过自主开发的LIF系统对不同温度下,花生油中AFB_(1)含量进行检测,探索温度对荧光信号的干扰规律。通过人为添加AFB_(1)标准品到10个品牌花生油中,配制不同AFB_(1)污染程度(控制组、0、5、10、20、25、30μg/kg)的花生油样本。使用LIF系统采集5种不同温度下(10、20、30、40、50℃)花生油的荧光光谱。光谱分析结果表明花生油荧光强度随着温度的升高而降低。使用5种不同建模方法建立全局模型,通过比较预测结果发现核函数为径向基函数(RBF)的支持向量机(SVM)方法所建立的全局模型预测效果最优,其正确率(CCR)超过99%,假阴性率(FNR)为0,假阳性率(FPR)为0.6%。在此基础上使用SVM(RBF)建立分温度模型,并使用各分温度模型分别对5个温度下验证集进行预测,结果表明各分温度模型在预测本温度验证集时,预测效果极优,正确率均高于97%,验证集FPR与FNR不超过5.5%,但在预测其他温度验证集时,效果极差。表明当使用LIF技术检测花生油AFB_(1)含量时,需要保证样本温度保持一致,或者通过建立温度全局模型提高模型鲁棒性。

Laser induced fluorescence spectroscopy(LIF)has been widely used in the detection of aflatoxin B_(1)(AFB_(1))in nuts,such as peanut and fig,etc.However,the fluorescence signal will be easily affected by environmental factors(mainly temperature).In the present study,the content of AFB_(1)in peanut oil at different temperatures was detected by the self-developed LIF system to explore the interference law of temperature on fluorescence signal.Different AFB_(1)contamination levels(control group,0,5,10,20,25,30μg/kg)were prepared by artificially adding AFB_(1)standard to 10 different brands of peanut oil.The fluorescence spectra of peanut oil at 5 different temperatures(10,20,30,40,50℃)were collected by LIF system.The results of spectral analysis indicated that the fluorescence intensity of peanut oil decreased with the increase of temperature.Five different modeling methods were used to establish the global model.Comparing the prediction results,it was found that,the global model established by the support vector machine(SVM)method with radial basis function(RBF)kernel achieved the best prediction results.Its accuracy(CCR)was more than 99%,false negative rate(FNR)was 0,and false positive rate(FPR)was 0.6%.On this basis,SVM(RBF)was used to establish the sub temperature model,and each sub temperature model was used to predict the verification sets under five temperatures.The results indicated that the prediction effect of each sub-temperature model was extremely good when the verification set was predicted under the corresponding temperature,the accuracy was higher than 97%,the FPR and FNR were less than 5.5%,but the results were very poor when predicting verification set under other temperature.It could be concluded that when LIF technology was used to detect the AFB_(1)content in peanut oil,it was necessary to ensure the consistency of sample temperature,or improve the robustness of the model by establishing a global temperature model.