基于激光诱导荧光光谱技术检测花生中黄曲霉毒素B1

Detection of aflatoxin B1 in peanuts based on laser induced fluorescence spectroscopy

中文关键词:  激光诱导荧光  荧光光谱  黄曲霉毒素B1  花生

英文关键词:laser induced fluorescence  fluorescence spectroscopy  aflatoxin B1  peanut

中文摘要:

      目的 探究激光诱导荧光(laser induced fluorescence, LIF)技术检测花生中黄曲霉毒素B1 (aflatoxin B1, AFB1)的可行性, 定性和定量分析花生中的AFB1。方法 制备不同浓度梯度的污染花生, 经LIF系统采集荧光光谱, 平滑后分析光谱数据结构。基于全波长光谱使用5种不同建模方法对污染花生定性判别, 采用偏最小二乘法回归(partial least squares regression, PLSR)和BP神经网络(BP neural networks, BPNN)进行定量预测。通过竞争性自适应重加权采样(competitive adaptive reweighted sampling, CARS)提取特征波长, 研究其对定性和定量预测的影响。结果 对于全波长光谱数据, 线性核函数的支持向量机[support vector machine with linear kernel function, SVM(Linear)]建立的判别模型效果最优, 预测正确率100.00%。PLSR和BPNN均获得较好的定量预测效果, 剩余预测偏差(residual predictive deviation, RPD)>3.0, 检出限(limit of detection, LOD)<20 μg/kg; 对于特征光谱数据, SVM(Linear)定性判别预测正确率93.94%, F1值为0.94, 受试者工作特征(receiver operating characteristic curve, ROC)曲线下面积(area under the curve, AUC)为0.989。建立的PLSR模型性能优于未提取特征波长的两种定量模型, RPD为3.36, LOD为14.76 μg/kg。结论 LIF技术检测花生中的AFB1简单快速, 定性定量预测模型准确性好, 具有一定可行性。

英文摘要:

      Objective To investigate the feasibility of the laser induced fluorescence (LIF) technique for the detection of aflatoxin B1 (AFB1) in peanuts, and to qualitatively and quantitatively analyze AFB1 in peanuts. Methods Different concentration gradients of contaminated peanuts were prepared, fluorescence spectra were collected by the LIF system, and the spectral data structure was analyzed after smoothing. Qualitative discrimination of contaminated peanuts based on full wavelength spectroscopy using five different modeling approaches, Partial least squares regression (PLSR) and BP neural networks (BPNN) were used for quantitative prediction. Feature wavelengths were extracted by competitive adaptive reweighted sampling (CARS) to investigate its effect on qualitative and quantitative prediction. Results For full wavelength spectral data, the discriminant model built by support vector machine with linear kernel function [SVM(Linear)] was the most effective, with a 100.00% correct prediction rate, and both PLSR and BPNN obtained better quantitative prediction results, the residual predictive deviation (RPD) was>3.0 and the limit of detection (LOD) was<20 μg/kg; for the feature spectral data, the SVM(Linear) qualitative discriminant predicted 93.94% correctly, F1 value was 0.94 and the area under the curve (AUC) of receiver operating characteristic curve (ROC) was 0.989. The performance of the established PLSR model was better than the two quantitative models without extracting the characteristic wavelengths, the RPD was 3.36 and the LOD was 14.76 μg/kg. Conclusion The detection of AFB1 in peanut by LIF technique is simple and rapid, and the qualitative and quantitative prediction model has good accuracy, which has certain feasibility.