基于可见光近红外反射光谱的土壤有机质含量研究.docxVIP

基于可见光／近红外反射光谱的土壤有 机质含量研究木 基于可见光／近红外反射光谱的土壤有 机质含量研究木 汪玉冰+，庄重，鲁翠萍，汪六三，宋良图，王儒敬，葛运建 (中国科学院，合肥智能机械研究所，安徽，合肥，230031) 摘要：本文采用可见光／近红外近红外光谱对土壤中有机质的含量进行实验研究，研究反射光谱的预处理方 法以提高光谱与有机质含量的相关性，并尝试使用不同回归方法构建光谱对有机质含量的预测模型，预测 未知样品土壤有机质含量。实验结果表明，对S-G多项式处理后的一阶导数和二阶导数进行主成分回归， 以及对原光谱前十个主成分的三层BP神经网络模型均能对土壤有机质含量做较好的预测。下一步研究中， 我们会逐步将仪器应用于田间，将机器运行噪声以及土壤颗粒大小、温湿度等实地因素考虑到模型搭建中， 从而真正地实现对土壤养分的实时准确预测。 关键词：土壤；有机质；近红外光谱；主成分回归；神经网络中图分类号：S151．9 Study of the soi l organi c matter content based on the VIS／NIR reflectance spectra Wang Yubing*，Zhuang Zhong，Lu Cuiping，Wang Liusan，Song Liangtu，Wang Ruj ing，Ge Yunjian(Institute of Intelligent Machines，Chinese Academy of Science， Hefei 230031，China) Abstract：In present work，the visible／near—infrared spectra have been used to study the content of soi 1 organic matter．Different pretreatment methods have been studied to improve the correlation of the spectra with soil organic matter content．Three different regression methods have been studied to construct the prediction model with the of VIR／NIR spectra，which be applied to predict organic matter content of unknown soil samples．The experimental results show that the Principal Component Regression for the first and the second derivative spectra by S-G polynomial，and the BP neutral network model for the first ten principal components of the original spectra both do good prediction of the soil organic matter content．In the fprther research，we will apply this instrument to the field operation gradually：moreover，we will take the machine noise and many practical factors such the size of soil particle，temperature humidity and on，into the predict ion model founding，in order to achieve real—t ime accurate prediction of sol 1 nutrients Key words：sol 1：organic matter：near infrared spectroscopy：principal component regression neural network 1．引言 精准施肥作为现代精准农业的一个重要组成部分，要求能根据不同地区、不 项目来源：中科院知识创新