地面LiDAR与高光谱数据配准及在单木结构参数提取中的应用

发布时间：2018-06-03 23:42

  本文选题：LiDAR + 高光谱　； 参考：《电子科技大学》2015年硕士论文

【摘要】：高速发展的遥感技术,使得传统光学遥感已经无法满足林木资源调查对空间信息和光谱信息的精度要求。近年来,由于激光雷达(Light Detection and Ranging,LiDAR)高精度的三维空间信息以及高光谱数据丰富的光谱信息,使二者迅速在各行各业得到了广泛应用。在LiDAR数据与高光谱数据空间信息与光谱信息互相补偿的条件下,协同两种数据将更有利于提高林木参数反演精度。本文利用地面三维激光扫描仪Leica Scanstation C10和成像光谱仪SOC710协同获取单木的三维空间信息和光谱信息,进行了以下几个方面的工作:(1)总结了国内外针对LiDAR数据和高光谱数据的研究现状,重点介绍了协同两种数据在林业方面的研究现状以及LiDAR与光学遥感数据配准方法的研究现状;(2)针对地面LiDAR点云数据的离散性和巨大的数据量,本文建立了基于四叉树的LiDAR点云数据索引机制;(3)针对地面LiDAR数据与高光谱数据成像方式的不同,LiDAR点云数据是离散的三维空间信息,高光谱数据是二维光谱图像,本文模拟相机成像方式,将LiDAR点云数据二维图像化;(4)针对LiDAR自身强度信息和高光谱数据灰度信息的差异,本文研究了基于控制点、基于特征以及基于互信息的配准方法,提出结合控制点和特征互信息的配准方法,利用基于控制点的配准方法实现LiDAR图像与高光谱图像的粗配准,然后将得到的粗配准参数作为Powell算法搜索的初始化参数,利用特征互信息作为相似性度量,该方法实现了LiDAR图像与高光谱图像的精配准。(5)针对利用地面LiDAR自身强度信息无法实现叶片与枝干分割,基于几何信息实现LiDAR数据的叶片与枝干分割,计算量非常大,本文根据树的叶片和枝干光谱曲线的区别,协同高光谱数据的光谱信息实现了单木LiDAR数据枝干与树叶的分割,最后利用树叶LiDAR点云数据,实现了基于VCP(Voxel-based Canopy)算法的叶面积密度估计。总之,本文针对地面LiDAR点云数据和高光谱数据的特点,充分利用基于控制点的图像配准方法、基于特征点的图像配准方法和基于互信息的配准方法,实现了基于控制点和特征互信息的地面LiDAR点云数据与高光谱图像的混合配准,最后协同地面LiDAR与高光谱图像估计单棵树的叶面积密度。
[Abstract]:With the rapid development of remote sensing technology, traditional optical remote sensing has been unable to meet the precision requirements of spatial and spectral information in forest resource survey. In recent years, because of the high precision three-dimensional spatial information and rich spectral information of LiDAR, both of them have been widely used in various industries. Under the condition that the spatial information and spectral information of LiDAR data and hyperspectral data are mutually compensated, it is more advantageous to improve the retrieval accuracy of tree parameters by cooperating the two kinds of data. In this paper, the 3D spatial and spectral information of a single tree is obtained by using Leica Scanstation C10, a 3D laser scanner, and the imaging spectrometer SOC710. In this paper, we have done the following work: 1) summarize the research status of LiDAR data and hyperspectral data at home and abroad. In this paper, the present situation of forestry research of two kinds of cooperative data and the research status of LiDAR and optical remote sensing data registration method are introduced in detail) aiming at the discreteness and huge amount of data of ground LiDAR point cloud data. In this paper, the indexing mechanism of LiDAR point cloud data based on quadtree is established. Aiming at the different imaging modes of ground LiDAR data and hyperspectral data, the point cloud data of LiDAR is discrete three-dimensional information, and the hyperspectral data is two-dimensional spectral image. In this paper, we simulate the camera imaging mode and transform the LiDAR point cloud data into two dimensional images. Aiming at the difference between the intensity information of LiDAR and the gray level information of hyperspectral data, the registration method based on control point, feature and mutual information is studied in this paper. A registration method combining control point and feature mutual information is proposed. The rough registration of LiDAR image and hyperspectral image is realized by using the registration method based on control point, and then the coarse registration parameters are used as initialization parameters searched by Powell algorithm. Using feature mutual information as similarity measure, the method realizes fine registration of LiDAR image and hyperspectral image. According to the difference of the spectral curves of the leaves and branches of the LiDAR data, the spectral information of the single tree LiDAR data can be segmented according to the difference of the spectral curves of the leaves and branches of the tree, and the spectral information of the hyperspectral data can be used to realize the segmentation of the branches and the leaves of the single tree, which is based on the geometric information. Finally, the leaf area density estimation based on the VCP(Voxel-based Canopy algorithm is realized by using the leaf LiDAR point cloud data. In a word, according to the characteristics of ground LiDAR point cloud data and hyperspectral data, this paper makes full use of the image registration method based on control point, the image registration method based on feature point and the registration method based on mutual information. The mixed registration of ground LiDAR point cloud data and hyperspectral images based on mutual information of control points and features is realized. Finally, the leaf area density of a single tree is estimated in collaboration with ground LiDAR and hyperspectral images.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TN957.52

【参考文献】

相关期刊论文 前1条

1 黄先锋,陶闯,江万寿,龚健雅;机载激光雷达点云数据的实时渲染[J];武汉大学学报(信息科学版);2005年11期

，

本文编号：1974775