基于高光谱数据的土壤含水量反演  

Inversion of Soil Moisture Content Based on Drone Hyperspectral Data

作　　者：陈潜 CHEN Qian(Meitan County Natural Resources Survey and Land Spatial Planning Center,Zunyi Guizhou 563000)

机构地区：[1]湄潭县自然资源调查与国土空间规划中心,贵州遵义563000

出　　处：《长江信息通信》2024年第12期19-22,共4页Changjiang Information & Communications

摘　　要：土壤含水量是农业生产发展的重要因素之一,利用无人机搭载高光谱仪可作为一种新型的快速、准确预测土壤含水量的科技手段,对我国农业发展具有重要意义。为消除无人机高光谱数据中背景噪声、冗余性、共线性问题,文章提出对光谱数据进行变换处理后进行主成分分析,将主成分变量作为模型输入变量建立BP神经网络反演模型(PCA-BPNN)和随机森林模型(PCA-RF)进行土壤含水量预测,并利用R2、RMSE、RPD的综合评价指标对两种反演结果进行精度验证与比较。结果表明:光谱倒数对数变换处理能够有效提高模型精度和预测能力,基于倒数对数光谱的PCA-RF模型精度最高(R_(M)^(2)=0.892,RPDV=1.474)。Soil moisture content is one of the important factors for the development of agricultural production.The use of unmanned aerial vehicles equipped with spectrometers can serve as a new scientific and technological means for quickly and accurately predicting soil moisture content,which is of great significance for the development of agriculture in China.In order to eliminate the background noise,redundancy and collinearity problems in UAV hyperspectral data,this paper proposes to transform the spectral data and conduct principal component analysis.The principal component variables are used as model input variables to establish BP neural network inversion model(PCA-BPNN)and random forest model(PCA-RF)to predict soil water content,and R2,RMSE The comprehensive evaluation index of RPD verifies and compares the accuracy of two inversion results.The results show that the spectral reciprocal logarithmic transformation processing can effectively improve model accuracy and prediction ability,and the PCA-RF model based on reciprocal logarithmic spectroscopy has the highest accuracy(R_(M)^(2)=0.892,RPD V=1.474).

关 键 词：高光谱 土壤水含量 主成分分析 

分 类 号：P237[天文地球—摄影测量与遥感]