基于无人机高光谱遥感的冬小麦叶片氮含量建模与反演研究

【作者】 刘海英；

【导师】 王萍；

【作者基本信息】 山东科技大学 ， 摄影测量与遥感， 2018， 博士

【摘要】 近年来,具有数据精度高、成像连续且方便快捷等特点的无人机高光谱遥感,成为国内外众多学者和科研机构的研究热点。基于无人机高光谱遥感的农作物定量参数的辐射传输机理分析与反演模型建立,是实现精准农业信息化的重要的理论和方法。本文以实现冬小麦叶片氮含量(LNC)无人机高光谱定量估算和遥感反演为研究目标,进行了无人机结合地面实测的高光谱遥感定量观测试验,获取了高光谱观测数据以及对应的农学生化参数数据,比对分析了无人机高光谱数据的光谱采样精度,获取了不同变量条件下的冬小麦冠层高光谱特征,提取了无人机高光谱数据中响应冬小麦LNC的有效代表波段,综合运用光谱指数法、多元统计回归、机器学习等方法,建立了冬小麦LNC的高光谱定量估算模型并进行了比较分析,最终实现了基于无人机高光谱的冬小麦LNC遥感定量反演和氮营养状况的初步诊断。论文主要在以下几个方面开展工作并取得了相应的研究成果:(1)基于地面实测和理论模型模拟,进行了无人机高光谱数据的比对分析基于与无人机观测(搭载UHD185高光谱成像仪)同步的地面实测(ASD高光谱仪)的高光谱数据,运用波段相关系数、光谱指数计算方法,对无人机高光谱数据的光谱采样精度进行了对比分析;同时,基于地面同步观测的冬小麦生化参数和PROSAIL植被辐射传输模型,获取了冬小麦冠层模拟反射光谱数据,利用模型模拟光谱数据对无人机高光谱数据进行了比对分析。两种比对结果表明,无人机高光谱数据具有较高的辐射分辨率和光谱采样精度,可以满足农作物遥感定量监测的精度和准确性要求。(2)细致分析了不同变量条件下的冬小麦冠层高光谱特征基于获取的无人机高光谱数据,分别从不同生育期,不同施氮水平、不同水分条件和不同叶片氮含量等四个方面,对冬小麦冠层高光谱特征进行了细致分析。分析结果表明,不同生育期的冬小麦的光谱反射率曲线总体趋势一致,但在不同的波段范围差异性显著,其中在近红外波段范围的差异大于可见光区域;随着施氮量、施水量和叶片氮含量增大,冬小麦在可见光区域的冠层光谱反射率减小,呈现负相关,而在近红外区域的光谱反射率则增高,呈现正相关,且近红外区域的变化幅度要大于可见光区域。该分析结果证实了在不同变量条件下,冬小麦冠层高光谱在特定波段上具有显著的“位移”和变异特征。(3)提出了一种基于波段相关性阈值分析的高光谱数据降维与代表波段选择方法,并提取了响应冬小麦不同生育期LNC的无人机高光谱特征波段基于高光谱数据谱段相邻愈近愈相关的思想,提出了一种利用光谱相关性并规定其阈值的高光谱波段降维和代表波段选择的方法;应用该方法和一阶微分、连续统去除的传统方法,分别提取了响应冬小麦不同生育期LNC的无人机高光谱代表波段;利用三种代表波段提取结果,分别进行了冬小麦不同生育期LNC的初步建模,分析比较三种建模结果发现:基于波段相关性阈值法得到的代表波段进行冬小麦LNC定量建模结果最优,提取的波段结果具有波段数目少、覆盖波谱范围广、波段代表性强的优点。该结果证实了基于波段相关性阈值进行高光谱数据降维与代表波段选择具有理想的效果。(4)基于波段相关性阈值获取的代表波段,构建了 LNC光谱指数,并进行了冬小麦不同生育期LNC的遥感估算。基于波段相关性阈值方法提取的响应冬小麦不同生育期LNC的代表性波段,分别构建了比值光谱指数、差值光谱指数、归一化光谱指数,并引入450nm蓝光波段对LNC光谱指数进行了修正;在此基础上,建立了基于LNC光谱指数的的冬小麦不同生育期LNC定量估算模型;与现有的氮光谱指数的建模结果相比,模型精度得到了较好的提高,其中挑旗期和开花期提高最为显著,同时,450nm修正波段的引入对于开花期和灌浆期的LNC光谱指数意义较大。(5)构建了冬小麦不同生育期LNC的无人机高光谱定量估算模型,实现了冬小麦LNC的遥感反演,并对冬小麦氮营养状况进行了初步的诊断与分析。基于波段相关性阈值分析提取的响应冬小麦不同生育期LNC的代表性波段,运用多元统计回归、随机森林、BP神经网络等方法,分别建立了冬小麦不同生育期LNC的无人机高光谱定量估算模型,并基于其他年份的地面实测数据对建模结果进行验证分析,得到了最优的冬小麦不同生育期LNC的无人机高光谱定量估算模型;应用最优的定量估算模型,进行了冬小麦LNC的无人机遥感反演:在此基础上,通过建立冬小麦LNC和氮营养指数之间的定量模型,实现了冬小麦不同生育期氮营养状况的无人机高光谱初步诊断与分析。本研究是无人机高光谱数据应用于农业定量参数估算与遥感反演的成功实践,设计提出的无人机高光谱数据比对分析和响应冬小麦LNC的代表波段提取方法,构建的冬小麦不同生育期LNC无人机高光谱定量估算模型,以及应用该模型实现的遥感反演,为无人机高光谱定量遥感及其应用提供了完整的方法思路,具有重要的理论和现实意义。更多还原

【Abstract】 Since the end of the 20th century,the unmanned aerial vehicle(UAV)remote sensing was rapidly rised and flourished,and it had becomed one of the leading aviation remote sensing technologies in current and future.So,the UAV remote sensing had broad prospect of application.In recent years,as the imaging hyperspectral remote sensing technology rapid development and wide application,the imaging hyperspectral remote sensing based on UAV had characteristics such as high accuracy,the imaging continuously and quickly.So,it had caused a research upsurge both in China and abroad.In this paper,the leaf nitrogen content of winter wheat was taken as the research object,and the ground plot test of winter wheat was carried out in 2014-2015.Near-surface winter wheat hyperspectral data was acquired using a drone-mounted imaging spectrometer and a ground-based non-imaging spectrometer.Combining with ground measurement data,the UAV spectroscopy characteristics of winter wheat were analyzed to extract the characteristic spectrum and sensitive spectral parameters of the response leaf nitrogen content of UAV in winter wheat.Comprehensively using multivariate regression,random forest,BP neural network and other methods,an estimation model for the leaf nitrogen content of winter wheat was established,and the optimal model was used to perform remote sensing inversion of the UAV hyperspectral imagery.On this basis,the relationship between leaf nitrogen content and nitrogen status of winter wheat was analyzed.An inversion model of nitrogen nutrient index of winter wheat was established and remote sensing image inversion was carried out to achieve a rapid diagnosis of nitrogen status of winter wheat.The paper mainly developed in following several aspects works and had obtained the innovation research results and the contribution:(1)Verified the effectiveness and availability of high spectral data obtained by UAVThe validity of hyperspectral data acquired by UHD185 sensor on UAV was verified.Based on the ground synchronous measured hyperspectral data,the accuracy and validity of UAV hyperspectral data was tested by using the correlation coefficient and spectral index calculation method.And based on the simulated reflectance spectrum data of PROSAIL vegetation radiation transmission model,two aspects of different growth period and different nitrogen application level were tested again.The results show that the UAV hyperspectral data had high data sampling accuracy and high reliability.(2)Analysis of UAV hyperspectral characteristics of winter wheat From different aspectsAccording to the UAV hyperspectral data,characteristics of canopy spectra of winter wheat were analyzed from four aspects:different growth stages,different nitrogen levels,different water conditions and different leaf nitrogen contents.The spectral characteristics of winter wheat at different growth stages were different,but the overall trend was the same.In the visible region,the canopy reflectance of winter wheat decreased slightly from jointing stage to flag leaf stage.From flag leaf stage to filling stage,the canopy reflectance of winter wheat increased continuously,and the reflection peak at 550nm became less,and tended to disappear.In the near infrared region,from jointing stage to flag leaf stage of winter wheat,canopy reflectance spectra showed an increasing trend;from flag leaf stage to flowering stage,canopy reflectance decreased gradually and tended to be stable;from the flowering stage to the filling stage,the spectral red edge position shifted to short wavelength,and the phenomenon of "blue shift"appeared.The UAV hyperspectral differences of winter wheat under different nitrogen levels were obvious.The near infrared region difference was more obvious than the visible region.The near infrared band had more sensitivity to nitrogen fertilizer application.With the increase of nitrogen application rate,the spectral red edge position shifted to long wavelength,and the phenomenon of "red shift" appeared.Under different water conditions,the difference of hyperspectral curves in the near infrared region of winter wheat canopy was greater than that in the visible region.In the visible range,with adequate irrigation water,the reflectance of Winter Wheat Canopy Spectra decreased,while in the near infrared region,winter wheat canopy spectral reflectance showed a rising phenomenon and the difference was significant.The results showed that the near infrared spectrum of canopy biomass of winter wheat was more obvious with the increase of soil water conditions.The spectral characteristics of Winter Wheat with different nitrogen content were different.In the visible region,the reflectance of the spectrum was negatively correlated with leaf nitrogen content.The lower the leaf nitrogen content was,the higher the spectral reflectance.In the near infrared region,the reflectance of the spectrum was positively correlated with leaf nitrogen content,and the reflectance increased with the increase of the leaf nitrogen content.(3)Proposed a hyperspectral representative band extraction method based on band correlation threshold in response to winter wheat leaf nitrogen contentAccording to the UAV hyperspectral data,the first order differential and continuous removal methods were used to extract the representative bands of LNC,and the results of the extraction band were modeled,analyzed and verified.A hyperspectral dimension reduction method was proposed based on correlation threshold analysis.Using this method,the representative bands combination of four key growth periods of LNC was obtained.Analyzed and compared the three results,which based on the hyperspectral correlation threshold was better,the number of bands was much less and the band was more representative.From the results of modeling,the R2 values of jointing stage,flag leaf stage and flowering stage were greatly improved.(4)New spectral indexes were constructed based on the band correlation threshold,the estimation of LNC of winter wheat at different growth stages of UAV hyperspectral period was achieved.The present spectral index was improved and optimized based on the representative band extracted by the band correlation threshold method in response to LNC of different growth stages of winter wheat.The ratio spectrum index,the difference spectral index and the normalized spectral index were constructed by the two combinations of the representative bands.On this basis,the 450nm band was introduced to modify the newly constructed spectral index.A new model for the LNC estimate in different growth stages was established based on the newly constructed spectral indexes.The spectral indexes constructed from different band combinations and sensitive bands were obtained in response to LNC in different growth stages.Compared with the existing modeling results of nitrogen spectrum index,the precision of the model had been improved,and the improvement of the flag leaf stage was the most obvious.The introduction of the 450nm correction band was of great help to the improvement of the modeling precision by the newly constructed spectral index during the flowering and filling stage.(5)Monitoring of LNC and preliminary diagnosis of nitrogen nutrition status of Winter Wheat based on UAV hyperspectral imagingBased on the representative bands extracted by the method of band correlation threshold,the hyperspectral quantitative model of UAV was established by multiple linear regression and random forest and BP neural network responding to LNC.The modeling results were compared and analyzed.Remote sensing image inversion is realized by selecting the optimal result model for remote sensing mapping.The change of nitrogen nutrition index of Winter Wheat under different nitrogen level was analyzed,and the estimation model of LNC and nitrogen nutrition index of winter wheat was established.The model was analyzed and verified,and remote sensing image inversion was carried out to realize the remote sensing monitoring of nitrogen nutrition status of winter wheat.This research has successfully carried out the data processing and application of UAV imaging hyperspectral data,and provided a method of UAV remote sensing image quantitative remote sensing application,which has important theoretical and practical significance.更多还原