土壤Cu含量高光谱反演的BP神经网络模型  

Hyper-spectral inversion of soil Cu content based on BP neural network model

作　　者：郭云开[1,2] 刘宁[1,2] 刘磊[1,2] 李丹娜 朱善宽[1,2] 

机构地区：[1]长沙理工大学交通运输工程学院,长沙410076 [2]长沙理工大学测绘遥感应用技术研究所,长沙410076

出　　处：《测绘科学》2018年第1期135-139,152,共6页Science of Surveying and Mapping

基　　金：国家自然科学基金面上项目(41471421;41671498)

摘　　要：以高光谱数据为基础,针对传统土壤重金属反演模型拟合度低、预测效果差的缺点,提取光谱预处理后的特征波段数据进行相关性分析,选取860nm一阶微分光谱反射率建立基于Matlab的重金属Cu含量BP神经网络预测模型,模型的拟合优度为0.721,预测精度达82.3%,高于传统单元线性回归模型0.414的拟合优度与76.1%的预测精度。研究表明,BP神经网络模型具有良好的拟合优度与预测能力,能更有效预测土壤中重金属Cu的含量。Based on the hyper-spectral data, for the reason of low fitting degree and poor prediction effect of the traditional inversion models of heavy models in soil, this paper extract the feature bands data of pretreated spectral for the correlation analysis, choose the first order differential spectral reflectance of 860 nm to establishes the BP neural network model of heavy metal Cu which based on Matlab, the fitting goodness of the model is 0. 721 and the prediction accuracy is up to 82.3%, which were higher than those of the traditional unit linear regression model＇s fitting goodness of 0. 414 and prediction accuracy of 76.1% the study show that BP neural network model has better goodness of fit and prediction ability to predict the content of heavy metal Cu in soil more effectively.

关 键 词：高光谱 土壤重金属 BP神经网络 单元线性回归 拟合优度 

分 类 号：TP753[自动化与计算机技术—检测技术与自动化装置]