水稻叶绿素含量高光谱反演模型及尺度转换方法研究

发布时间：2018-05-04 11:38

本文选题：叶绿素含量 + BP神经网络　；参考：《中国地质大学(北京)》2015年硕士论文

【摘要】：叶绿素是作物在进行光合作用时的一类重要的色素,直接控制着作物能量传递和物质循环过程,其含量的变化可以用来评价作物光合作用的能力、作物受重金属污染胁迫的程度、以及作物的营养水平等。因此,作物叶绿素含量的监测对于农业的生产具有重要意义。本研究以长春的4块水稻田作为采样区域,通过地面实测的ASD光谱数据和实测的水稻叶绿素含量数据,建立水稻叶绿素含量的高光谱反演模型。通过分析水稻光谱曲线的“峰-谷”形态特征,利用Hyperion影像,将已建立的模型进行升尺度,并在区域尺度进行应用。本文主要工作及结论如下:(1)本文建立了基于BP神经网络算法的水稻叶绿素含量高光谱反演模型,其输入变量为根据前人的研究成果和试验分析得到的4个植被指数,隐含层个数为1,含有12个神经元节点,输出变量为叶绿素含量数据,模型结构为4-12-1。决定系数R2=0.882,均方根误差RMSE=2.958。(2)通过对已建立的水稻叶绿素含量高光谱反演模型的分析比较得到,基于BP神经网络算法建立的水稻叶绿素含量高光谱反演模型与其它的统计模型(基于原始光谱及其变体的各自的一阶导数和植被指数的单变量和多变量统计模型)相比,其决定系数较高,均方根误差值较低,能够更好地对水稻叶绿素含量进行监测。(3)研究发现,水稻的光谱曲线存在明显的“峰-谷”形态特征,ASD实测光谱数据及Hyperion影像像元反射率数据,均可以表示为以波长λ为自变量,反射率数据为因变量的分段函数,且两条曲线之间形状相似,由数学知识我们可得,两条曲线可以通过曲线的伸缩平移来进行相互转换。(4)本研究根据以光谱“峰-谷”形态特征为依据的分段函数,建立ASD实测光谱反射率数据和Hyperion影像像元反射率数据之间的线性转换关系,通过模型的尺度转换,来进行水稻叶绿素含量的区域监测,达到利用Hyperion遥感影像进行快速、无损的水稻叶绿素含量遥感监测的目的。
[Abstract]:Chlorophyll is a kind of important pigments in photosynthesis, which directly control the energy transfer and material cycle of crops. The change of chlorophyll content can be used to evaluate the ability of crop photosynthesis. The degree of heavy metal pollution to crops and the nutrient level of crops. Therefore, the monitoring of crop chlorophyll content plays an important role in agricultural production. In this study, the hyperspectral inversion model of rice chlorophyll content was established by using four paddy fields in Changchun as sampling area, and based on the measured ASD spectral data and the measured chlorophyll content data of rice. By analyzing the "peak-valley" morphological characteristics of the spectral curve of rice, the established model was scaled up by using Hyperion image and applied to the regional scale. The main work and conclusions of this paper are as follows: (1) in this paper, a hyperspectral inversion model of rice chlorophyll content based on BP neural network algorithm is established. The input variables are four vegetation indices based on the previous research results and experimental analysis. The number of hidden layers is 1, there are 12 neuron nodes, the output variables are chlorophyll content data, and the model structure is 4-12-1. Based on the analysis and comparison of the established hyperspectral inversion model of chlorophyll content in rice, the determination coefficient is 0.882and the root mean square error (RMSE) is 2.958.2. The hyperspectral inversion model of rice chlorophyll content based on BP neural network algorithm is compared with other statistical models (univariate and multivariate statistical models based on the first derivative of the original spectrum and its variants and the vegetation index). Its determination coefficient is higher, the root mean square error is lower, and it can better monitor the chlorophyll content of rice. There are obvious "peak-valley" morphological characteristics in the spectral curve of rice. The measured spectral data and the pixel reflectivity data of Hyperion images can be expressed as piecewise functions with wavelength 位 as independent variable and reflectivity data as dependent variable. And the shapes of the two curves are similar, and we can get from the mathematical knowledge that the two curves can be converted to each other by means of the stretching and translation of the curves.) the present study is based on the piecewise functions based on the morphological characteristics of the spectral "peak-valley". The linear conversion relationship between spectral reflectance data measured by ASD and pixel reflectance data of Hyperion image was established. The regional monitoring of rice chlorophyll content was carried out by scale conversion of the model, and the rapid use of Hyperion remote sensing image was achieved. Objective of remote sensing monitoring of chlorophyll content in rice.
【学位授予单位】：中国地质大学(北京)
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：S511;S127

【参考文献】