高光谱土壤有机质估测模型对比研究  

Contrast research on soil organic matter estimation model using hyper-spectral data

作　　者：袁征[1] 李希灿[1] 于涛[1] 张广波[1] 

机构地区：[1]山东农业大学信息科学与工程学院,山东泰安271018

出　　处：《测绘科学》2014年第5期117-120,164,共5页Science of Surveying and Mapping

基　　金：国家科技计划项目(2011BAD21B0601)

摘　　要：应用高光谱技术探讨土壤有机质含量定量估测方法,对发展精细农业具有重要意义。本文利用陕西省横山县的实测数据,采用对数的一阶微分变换方法对土样的高光谱数据进行处理,分别采用线性回归分析法、BP神经网络法、模糊识别法建立高光谱土壤有机质含量估测模型,并对比分析其精度,确定最优的光谱反演模型。实验结果表明:模糊识别模型的决定系数达到0.973,RMSE为0.0468%;比线性模型和BP神经网络模型精度都高。研究表明,土壤有机质光谱反演不仅要重视机理研究,同时要加强光谱反演建模方法创新。Using hyper-spectral technology to explore the quantitative estimation of soil organic matter content is of great significance for developing precision agriculture.Firstly,according to the measured data in Hengshan county,Shanxi province,the paper used one-order differential of logarithm to transform the collecting soil hyper-spectral data,and then respectively used linear regression analysis,fuzzy recognition method and artificial neural network to establish hyper-spectral soil organic matter estimation model,finally through the contrast analysis on the precision,the optimal spectrum inversion model was found out.The results showed that in the fuzzy recognition model,R2 is 0.973,and RMSE is 0.0468%,better than in both linear regression model and BP artificial neural network model.Therefore,soil organic matter hyper-spectral estimation should not only be absorbed in spectra mechanism research but also strengthen modeling methods innovation.

关 键 词：土壤有机质 高光谱 模糊识别 估测模型 

分 类 号：P237.9[天文地球—摄影测量与遥感]