基于非线性径向基核函数支持向量机的堆肥产品近红外光谱分析研究  

Near Infrared Reflectance Spectroscopy Analysis of Compost Products Using Nonlinear Support Vector Machine With RBF Nucleus

作　　者：黄光群[1] 韩鲁佳[1] 

机构地区：[1]中国农业大学工学院,北京100083

出　　处：《光学学报》2009年第12期3556-3560,共5页Acta Optica Sinica

基　　金：国家"十一五"科技支撑项目(2006BAD10B05;2006BAD07A13)资助课题

摘　　要：探讨了径向基核函数（RBF）支持向量机（RBFSVR）建立定量分析模型时主要参数的优选方法及其在近红外光谱分析畜禽粪便堆肥产品含水率、挥发性固体含量和碳氮比中的应用，并与偏最小二乘回归法所建近红外定量分析模型的预测能力做了比较。供试样品为我国22省市不同种类的120个畜禽粪便堆肥产品样品，利用傅里叶变换型光谱仪获取样品在4000～10000cm^-1内的光谱数据信息。研究发现，逐步寻优循环优选支持向量机建模参数方法具有较好的可行性，其所建近红外定量分析模型均优于基于偏最小二乘法所建模型，所建立的含水率和挥发性固体近红外模型验证决定系数（r^2）均大于0．90，相对分析误差（RPD）均大于4．0，具有实际应用价值；所建碳氮比近红外模型验证决定系数（r^2）为0．85，RPD值大于2．5，也可用于定量分析，但精度有待于进一步提高。This study explored a new method to choose optimal parameters for support vector machine regression with RBF nucleus （RBF-SVR） and its application on the estimation of moisture content, volatile solid （VS） and the ratio of carbon to nitrogen （C/N） in animal manure compost products using near-infrared reflectance spectroscopy （NIRS）. The efficiency of RBF-SVR method was compared with partial least-squares regression （PLSR） mainly using the determination coefficient of prediction （ r^2 ） of the standard error of prediction （SEP） and ratio of porformance to standard deviation ERPD （SD/SEP）]. In this study, 120 commercial animal manure compost samples were collected from 22 provinces in China. Spectra of the orient samples were scanned with a SPECTRUM ONE NTS from 4000- 10000 cm^-1, respectively. Results showed stepwise search for optimal parameters was a feasible method for RBF-SVR. The efficiency of RBF-SVR method for moisture content, VS and C/N were all better than PLSR. Robust models using RBF-SVR were developed for moisture content and VS （ r^2 〉 0.90, RPD〉 4.0） and for C/N （ r^2 0.85, RPD〉 2.5）, respectively. Results showed the potential of NIRS with RBF-SVR to evaluate the products quality of animal manure compost, but further research would be needed for the higher precision.

关 键 词：光谱学 堆肥品质分析 近红外漫反射光谱 支持向量机 

分 类 号：S14[农业科学—肥料学] X7[农业科学—农业基础科学]