基于高光谱成像技术的全麦粉中DON含量预测方法研究  

Method Development for Detection of DON Content in Wholewheat Flour Based on Hyperspectral Imaging Technology

作　　者：邢常瑞 赵双玲 孙钰莹 张洁 李光磊 赵小旭 袁建 鞠兴荣 Xing Changrui;Zhao Shuangling;Sun Yuying;Zhang Jie;Li Guanglei;Zhao Xiaoxu;Yuan Jian;Ju Xingrong(College of Food Science and Engineering,Nanjing University of Finance and Economics,Jiangsu Modern Grain Circulation and Safety Collaborative Innovation Center,Jiangsu Key Laboratory of Grain and Oil Quality Control and Deep Processing Technology,Nanjing 210023;Henan Science and Technology Education Center of Agriculture and Rural,Zhengzhou 450008;Shandong Meizheng Biotechnology Co.,Ltd.,Rizhao 102100)

机构地区：[1]南京财经大学食品科学与工程学院,江苏省现代粮食流通与安全协同创新中心,江苏省粮油品质控制及深加工技术重点实验室,南京210023 [2]河南省农业农村科技教育中心,郑州450008 [3]山东美正生物科技有限公司,日照102100

出　　处：《中国粮油学报》2024年第8期82-88,共7页Journal of the Chinese Cereals and Oils Association

基　　金：国家重点研发计划项目(2022YFD2100204、2022YFD2100202);江苏高校优势学科建设工程资助项目(苏政办发〔2018〕87号)。

摘　　要：DON是小麦中检出率较高、危害较严重的真菌毒素之一,是小麦减产和品质劣变的主要威胁。本研究首先采集全麦粉中DON含量的短波红外高光谱信息,建立全麦粉中DON含量的预测模型,并采用连续投影算法(SPA)进行特征波长的选取,比较了全波长范围和特征波长下建立的偏最小二乘法(PLS)和支持向量机(SVM)模型。结果表明,基于全波长数据建立的全麦粉中DON含量的最优预测模型为SVM模型,对应的预测集决定系数R^(2) P为0.640,预测集均方根误差(RMSEP)为1904.43μg/kg,剩余预测残差(RPD)为2.00。基于特征波长建立的最优预测模型为Autoscale-SVM模型(SPA-Autoscale-SVM),模型对应的R^(2) P、RMSEP和RPD分别为0.716、1640.41μg/kg和2.06。预测值与测量值的回归系数R^(2)为0.7449,说明基于特征波长的Autoscale-SVM模型能够预测全麦粉中DON含量的变化。为验证所建模型的稳定性,重新挑选小麦样品进行高光谱图像采集,将独立验证集带入所建模型中,验证集的拟合回归系数为0.7178,说明该模型能对全麦粉中DON含量进行预测。DON is one of the mycotoxins with the highest detection rate and the most serious harm in wheat,is the main threat to wheat yield reduction and quality deterioration.In this study,the short-wave infrared hyperspectral information of DON content in wholewheat flour was first collected to establish a prediction model of DON content in whole wheat flour,and the characteristic wavelength was selected by continuous projection algorithm(SPA).The partial least square method(PLS)and support vector machine(SVM)models were established based on the full wavelength range and characteristic wavelength and compared.The results indicated that the optimal prediction model for DON content in whole wheat flour was SVM model based on the whole wavelength range.The corresponding prediction set determination coefficient R^(2) P was 0.640,the root mean square error of prediction set(RMSEP)was 1904.43μg/kg,and the residual prediction residual(RPD)was 2.00.The optimal prediction model based on the characteristic wavelength was Autoscale-SVM model(SPA-Autoscale-SVM),and the R^(2) P,RMSEP and RPD corresponding to the model were 0.716,1640.41μg/kg and 2.06,respectively.The regression coefficient R^(2) between the predicted value and the measured value was 0.7449,indicating that the SPA-Autoscale-SVM model could predict the change of DON content in whole wheat flour.In order to verify the stability of the established model,wheat samples were selected again for hyperspectral image acquisition,and the independent verification set was introduced into the established model.The fitting regression coefficient of the verification set was 0.7178,indicating that the model could predict DON content in whole wheat flour.

关 键 词：高光谱成像 呕吐毒素 小麦 PLS SVM 

分 类 号：S513[农业科学—作物学]