基于无人机多光谱的华北平原花铃期棉花叶片SPAD建模方法研究  

Modeling Method of Cotton Leaves SPAD at Flowering and Boll Stage in North China Plain Based on UAV Multi-Spectrum

作　　者：纪伟帅 陈红艳[1] 王淑婷 张玉婷 Ji Weishuai;Chen Hongyan;Wang Shuting;Zhang Yuting(School of Resources and Environment,Shandong Agricultural University,Tai'an Shandong 271018;Dongying Natural Resources and Planning Bureau Kenli Branch,Dongying Shandong 257500)

机构地区：[1]山东农业大学资源与环境学院,山东泰安271018 [2]东营市自然资源和规划局垦利分局,山东东营257500

出　　处：《中国农学通报》2021年第22期143-150,共8页Chinese Agricultural Science Bulletin

基　　金：山东省自然科学基金“基于星-机-地多源数据融合的滨海盐渍化耕地盐分含量反演模型”(ZR2019MD039)。

摘　　要：华北平原地区棉花叶片SPAD光谱特征有待探明,其最适宜建模方法亦有待研究。笔者针对华北平原棉区,基于无人机多光谱探索其叶片SPAD光谱特征和最佳建模方法。以德州市夏津县大李庄棉区为研究区,利用无人机获取棉花花铃期的多光谱图像,同步测定棉花叶片SPAD值;对原始光谱进行预处理并组合构建光谱指数,进而采用相关分析筛选出6个棉花SPAD特征光谱指数;分别采用BP神经网络(BPNN)、多元逐步回归(MSR)和支持向量机(SVM)方法构建棉花SPAD值定量分析模型,并对模型验证、对比,优选最佳模型和建模方法,进而定量分析研究区棉花叶片SPAD空间分布。结果表明:棉花叶片SPAD的特征波段为红光和红边波段;入选模型的特征光谱指数为r、r^(*)reg、(reg-r)/(reg+r)、r-g、r/g、√r^(2)+g^(2);对比3种建模方法,BPNN模型精度最高,其建模集R2、RMSE分别为0.747、4.568,验证集R^(2)、RMSE、RPD分别为0.758、4.142、2.135,确定为棉花叶片SPAD的最佳模型。基于BP神经网络模型进行棉花叶片SPAD的空间分布反演,反演值与实测值具有高度一致性,拟合结果较好。BP神经网络可以作为基于无人机多光谱的华北平原棉花叶片SPAD建模的优选方法,该研究可促进棉田定量遥感和棉花长势监测。SPAD spectral characteristics of cotton leaves in North China Plain need to be ascertained,and the most suitable modeling method is yet to be studied.Aiming at the cotton area of North China Plain and based on unmanned drone multi-spectrum,this paper explored SPAD spectral characteristics of cotton leaves and the best modeling method.Focusing on the cotton area of North China Plain in the Yellow River Basin,we took Dali village cotton area of Xiajin County,Dezhou City as research area,used unmanned drone to obtain multispectral images at flowering and boll stage,and simultaneously determined SPAD value of cotton leaves.In this paper,the original spectrum was preprocessed and combined to construct the spectral index,and then 6cotton SPAD characteristic spectral indexes were screened out by correlation analysis.BP neural network(BPNN),multiple stepwise regression (MSR) and support vector machine (SVM) methods were used respectively to construct quantitative analysis model of SPAD value of cotton,and verified,compared and optimized the best model and modeling method,and then quantitatively analyzed the spatial distribution of cotton leaf SPAD in the research area.The results show that:the characteristic bands of cotton leaf SPAD are red band and red edge band.The characteristic spectral index of the selected model are r,r^(*)reg,(reg-r)/(reg+r)r-g,r/g and √r^(2)+g^(2).Compared the three modeling methods,BPNN model has the highest accuracy.Its modeling set R2and RMSE are 0.747 and 4.568,respectively,and its verification set R^(2),RMSE and RPD are0.758,4.142 and 2.135,respectively,and the model is determined as the best one of cotton leaf SPAD.Based on BP neural network model,the spatial distribution of cotton leaf SPAD is inverted,and the inversion value is highly consistent with the measured value,and the fitting result is good.BP neural network could be used as a preferred method for SPAD modeling of cotton leaves of North China Plain based on unmanned drone multispectrum.This research could promote quantitative

关 键 词：棉花 SPAD 无人机多光谱 遥感 华北平原 

分 类 号：S127[农业科学—农业基础科学]