基于遥感数据的低山丘陵区苹果树冠层叶绿素含量反演

发布时间：2018-01-17 19:15

本文关键词：基于遥感数据的低山丘陵区苹果树冠层叶绿素含量反演　出处：《山东农业大学》2017年硕士论文　论文类型：学位论文

【摘要】：叶绿素是作物进行光合作用的主要载体,是检测作物光合作用能力和生长发育状况的重要指标。传统的叶绿素含量实验室化学测定方法,费时、费力,不利于大面积监控作物长势状况。而近年来快速发展的遥感技术以其监测速度快、成本低、面积大等优点,为作物叶绿素含量的反演提供了一种新的技术方法。因此,基于遥感技术反演作物叶绿素含量具有重要的理论与现实意义。本研究以“苹果之都”之称的山东省栖霞市为研究区,以Sentinel-2A遥感影像和近地面实测的苹果树冠层高光谱数据为数据源,遥感反演苹果树冠层叶绿素含量。首先,利用余弦校正和Minnaert模型,对大气校正后的遥感影像进行地形辐射校正;在结合近地面实测苹果树冠层高光谱数据进行混合像元分解的基础上,进行苹果树冠层反射率反演;然后,在借鉴前人已构建的植被指数的基础上,以Sentinel-2A影像的蓝光、绿光、红光、红边与近红外波段数据构建的植被指数,筛选苹果树冠层叶绿素植被指数;最后,基于植被指数构建苹果树冠层叶绿素含量的反演模型并进行检验,对比分析多种模型的精度,优选出最佳的反演模型。主要研究结果如下:(1)进行了苹果树冠层反射率反演及精度分析对Sentinel-2A多光谱遥感影像进行了大气校正,在此基础上,使用余弦校正和Minnaert模型对研究区进行了地形辐射校正。其中,Minnaert模型校正后,影像的均值和标准差均小于余弦校正后影像,其影像的均值接近于大气校正影像的均值,很好地去除了地形阴影,降低了阴阳坡对比度,消除或减弱了地形的影响,得到了地表反演反射率。结合近地面实测数据,利用线性模型对地表反演反射率影像进行了混合像元分解,得到了苹果树冠层的反演反射率。通过对影像进行处理,表观反射率、地表反演反射率、冠层反演反射率与冠层实测反射率的相对误差是逐步降低的。冠层反演反射率的数值和冠层实测反射率的值是最为相近的,波段2~8A的相对误差为14.4%、14.6%、9.5%、10.1%、1.6%、0.4%、1.4%和2.0%,说明通过各种影像处理得到了更加真实的冠层光谱,为后续分析提供了精度保证。(2)构建及筛选了苹果树冠层叶绿素植被指数通过综合考虑绿色植被的光谱特性及Sentinel-2A影像的波段,借鉴RVI、CI、NDVI的构造原理及形式,以Sentinel-2A的蓝光波段2、绿光波段3、红光波段4、红边波段7、近红外波段8和近红外波段8A构建了12种植被指数,通过与叶绿素含量进行相关性分析,并对植被指数进行自相关性分析,优选出了3个植被指数系列,为系列1(RVIblue+RVIred+RVIre)、系列2(CIblue+CIred+CIre)和系列3(NDVIgreen+NDVIred+NDVIre)。(3)建立与验证了苹果树冠层叶绿素含量反演模型以植被指数系列1、系列2和系列3分别为自变量,苹果树冠层叶绿素含量为因变量,建立了BP神经网络反演模型和支持向量机回归反演模型。以NDVIgreen+NDVIred+NDVIre植被指数建立的BP神经网络反演模型3的建模及检验的决定系数(Rc2=0.674,Rv2=0.601)均大于BP神经网络反演模型1和模型2,均方根误差(RMSEc=0.169,RMSEv=0.185)都小于模型1和模型2,反演效果比较好。以NDVIgreen+NDVIred+NDVIre植被指数建立的支持向量机回归反演模型3的建模及检验的决定系数均大于支持向量机回归反演模型1和模型2,分别为0.729和0.667,均方根误差都小于模型1和模型2,分别为0.159和0.178,反演效果比较好。支持向量机回归反演模型3优于BP神经网络反演模型3,表明支持向量机回归反演模型3效果最佳,可以很好地反演苹果冠层叶绿素含量,也表明Sentinel-2A影像在冠层叶绿素反演中的有效性。综上所述,Sentinel-2A遥感影像结合近地面高光谱测定数据,为低山丘陵区苹果树冠层叶绿素含量的宏观监测与快速诊断提供了新的方法,为农业信息化的发展提供了理论依据和技术支撑。
[Abstract]:Chlorophyll is the main carrier for crop photosynthesis, is an important indicator of the ability to detect crop photosynthesis and growth status. The chlorophyll content of laboratory chemical measuring method, the traditional time-consuming, laborious, not conducive to large-scale monitoring of crop growth status. The remote sensing technology in recent years the rapid development of the monitoring of fast speed, low cost, the advantages of the area so, it provides a new technical method for the inversion of crop chlorophyll content. Therefore, the chlorophyll content of crops based on remote sensing technology has important theoretical and practical significance. Based on the "Apple Capital" of the Shandong city of Qixia Province as the study area, the apple tree canopy hyperspectral remote sensing image data Sentinel-2A and near ground data as the data source, the remote sensing inversion of Apple Tree Canopy Chlorophyll content. Firstly, using cosine correction and Minnaert model, after atmospheric correction. The sense of image topographic correction; spectral unmixing based on the combination of near ground measurement of apple tree canopy hyperspectral data, of apple tree canopy reflectance inversion; then, on the basis of previous vegetation index has been constructed on the Sentinel-2A image of the blue, green, red, red and near-infrared vegetation index band data construction and screening of Apple Tree Canopy Chlorophyll vegetation index; finally, test the inversion model of vegetation index of chlorophyll content in apple tree canopy and based on the comparative analysis of various model precision, select the best inversion model. The main results are as follows: (1) the inversion and accuracy analysis of the reflectance of apple tree canopy Sentinel-2A multi spectral remote sensing image of atmospheric correction, on this basis, using the cosine correction and Minnaert model for topographic correction in the study area. Among them, Minnaert model after correction, the mean and standard deviation were less than the cosine correction image after image, the image of the mean close to mean in atmospheric correction of images, removing all terrain shadows, reduces the slopes of contrast, eliminate or weaken the influence of the topography, the surface reflectance inversion combined with near. The measured data, the surface reflectance inversion image of mixed pixel decomposition based on linear model, the inversion reflectance of apple tree canopy. Through processing the image, apparent reflectance, surface reflectance inversion, the relative error of canopy reflectance and canopy reflectance inversion is gradually reduced. Numerical and canopy reflectance inversion of canopy the reflectance values are most similar, the relative error of 2~8A bands was 14.4%, 14.6%, 9.5%, 10.1%, 1.6%, 0.4%, 1.4% and 2%, that obtained through various image processing The more real canopy spectra, ensure accuracy provided for subsequent analysis. (2) the construction and screening of Apple Tree Canopy Chlorophyll vegetation index by considering the spectral characteristics of Sentinel-2A image and the green vegetation of the band, from RVI, CI, construction principle and form of NDVI, with the blue band 2 of Sentinel-2A, the green band 3 red, red edge band 4, band 7, band 8 near infrared and near infrared 8A to construct 12 vegetation indices, the correlation analysis and the content of chlorophyll, and the vegetation index correlation analysis, selected 3 vegetation index series, series 1, series 2 (RVIblue+RVIred+RVIre) (CIblue+CIred+CIre) and 3 Series (NDVIgreen+NDVIred+NDVIre). (3) establishment and verification of the Apple Tree Canopy Chlorophyll Content Retrieval Model Based on Vegetation Index Series 1, series 2 and 3 respectively as independent variable and the Apple Tree Canopy Chlorophyll Content As the dependent variable, established the inversion model of BP neural network and support vector machine regression model. The coefficient of determination BP neural network inversion model based on NDVIgreen+NDVIred+NDVIre vegetation index model and inspection of the 3 (Rc2=0.674, Rv2=0.601) were more than BP neural network inversion model 1 and model 2, the root mean square error (RMSEc=0.169, RMSEv=0.185) are less than the model 1 and model 2, better retrieval results. Support vector machine based on NDVIgreen+NDVIred+NDVIre vegetation index regression model 3 model and inspection decision coefficient is greater than the support vector machine regression model 1 and model 2, respectively 0.729 and 0.667, the root mean square error is less than model 1 and model 2, respectively. 0.159 and 0.178, the inversion result is good. The support vector machine regression model 3 is better than BP neural network inversion model 3, show that the support vector machine regression model 3 Effect Good, can be a good inversion of Apple Canopy Chlorophyll content, also show the effectiveness of Sentinel-2A image in Canopy Chlorophyll inversion. In summary, near ground hyperspectral data with the Sentinel-2A remote sensing image, which provides a new method for low mountain and hilly area of Apple Tree Canopy Chlorophyll content of the macro monitoring and rapid diagnosis, provides the theory the basis and technical support for the development of agricultural information.

【学位授予单位】：山东农业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：S661.1;S127

【参考文献】