摘要
本研究基于高光谱技术和化学计量学方法,对薄荷叶上的异丙甲草胺、烟嘧(莠去津、敌草胺和砜嘧)精喹4类除草剂残留进行种类判别。高光谱成像光谱范围为450nm~950nm的可见-近红外区域。为降低噪音对光谱数据的干扰、提升判别精度,采用SG平滑和多元散射校正对高光谱曲线进行处理。利用主成分分析算法(PCA)对原始数据进行降维后,建立以径向基函数(RBF)函数为核函数的支持向量机(SVM)模型。分别利用网格搜索(GS)、粒子群算法(PSO)及灰狼算法(GWO)对SVM模型参数进行寻优,对比不同模型的判别正确率,并利用精确度、召回率、约登指数、ROC和AUC对模型判别和泛化能力进行评估。实验结果表明,SG-PSO-SVM、SG-PCA-GWO-SVM和MSC-GS-SVM对测试集的判别正确率达到了100%,其中SG-PCA-GWO-SVM计算时间最短,而MSC-GS-SVM具有最优的泛化能力,从而实现了对薄荷叶片上常见除草剂的快速无损识别。
In this study,a hyperspectral method was established to analyze the herbicide residues of Metolachlor,Nicosulfuron&Atrazine,Napropamide and Rimsulfuron&Quizalofop-P on mint leaves.The spectra ranged from 450nm to 950nm.In order to reduce the interference of noise on spectral data and improve the discrimination accuracy,the hyperspectral curves were processed by SG smoothing and multiple scattering correction..The dimension of original data were reduced by principal component analysis(PCA),a support vector machine(SVM)model with radial basis function(RBF)as kernel function was established.The grid search(GS),particle swarm optimization(PSO)and gray wolf optimizer(GWO)were used to optimize the SVM models parameters,and the discrimination accuracy of different models were compared.The precision,recall rate,Youden’s index,ROC and AUC were used to evaluate the models discrimination and generalization ability.The experimental results show that the accuracy of SG-PSO-SVM,SG-PCA-GWO-SVM and MSC-GS-SVM is 100%.SG-PCA-GWO-SVM has the shortest computation time,while MSC-GS-SVM has the best generalization ability.
作者
赵昱萱
黄威
董林沛
任昕昕
李开开
ZHAO Yu-xuan;HUANG Wei;DONG Lin-pei;REN Xin-xin;LI Kai-kai(College of Criminal Investigation,National Experimental Teaching Demonstrating Center,People’s Public Security University of China,Beijing 100038,China;Institute of Forensic Science,Ministry of Public Security,Beijing 100038,China)
出处
《化学研究与应用》
CAS
CSCD
北大核心
2022年第1期91-102,共12页
Chemical Research and Application
基金
国家重点研发计划项目(2019YFF0303405)资助
中央高校基本科研业务费专项资金项目(2021JKF201)资助。
关键词
除草剂残留
高光谱图像
化学计量法
判别分析
无损检验
herbicide residues
hyperspectral imaging
stoichiometry
discriminant analysis
non-destructive
