基于线特征的国产高分影像几何校正方法研究

论文总字数：82493字

摘 要

遥感影像不仅在城市变迁、道路扩建、植被变化以及土壤侵蚀等方面可以反映出一个地区的动态变化信息，还可以在军事上帮助军方快速找到城镇、机场、道路以及桥梁等重要信息。但是由于遥感影像在成像过程中会发生几何畸变，只有经过精确几何校正的遥感图像才能在各方面发挥更大的作用。在我国西北高寒山区等特殊地貌背景下，同名控制点获取十分困难，导致常规基于控制点的几何校正方法实施困难。

当前GNSS发展迅速，车载导航仪轨迹数据越来越丰富，且获取便捷；同时高分影像上有较多线状地物，采用线特征提取技术可以获取影像线状地物中心线，本文以南京地区GF-1影像为数据源，研究了一种基于线特征的影像几何校正方法。首先，利用遥感技术提取高分影像上线状物的中心线，同时实测这些线状物的GNSS轨迹；其次，通过线特征描述构建影像线状物和实测GPS轨迹的线特征集，基于线特征匹配算法获取同名线对，并基于同名线建立校正模型，再利用少量确定的同名控制点对模型进行优化；最后依据校正模型对影像实施几何校正，并对影像校正结果做出精度评价。本文通过研究与分析得到以下结论：

（1）基于C4.5算法在道路中心线提取的过程中提高了智能化和自动化程度，由实验结果表明该方法对简单道路提取结果的正确度为90%，较复杂的路网提取结果正确度为87%，中误差为。

（2）通过影像中线状地物中心线和实测GNSS轨迹线的空间结构和几何关系进行同名线匹配，可以快速、准确的获取同名线对。由实验结果表明，实验GF-1影像中同名线对之间的距离阈值取31m、矩形宽度阈值取65m、斜率比率阈值取0.4时，匹配的正确率最高。

（3）基于线特征的校正模型可以解决影像的尺度、旋转和平移问题，再通过少量确定同名点选择一定量的控制点进行多项式优化校正，可有效提高校正精度。最终实验结果表明，该方法校正后的中误差在以内，由此证实了在缺少控制点的情况下，基于线特征建立校正模型对影像进行校正方法的可实施性。

关键词：信息提取，线特征匹配，几何校正，高分影像

Study on Geometric Correction Method of Domestic High-resolution Image Based on Line Features

Abstract

Remote sensing image can not only reflect the dynamic transformation information of an area in the aspects of urban change, road expansion, vegetation change and soil erosion, but also can help the military to quickly find important information such as towns, airports, roads and bridges. However, due to the distortion of the remote sensing image in the imaging process, only the remote sensing image with accurate geometric correction can play a greater role in all aspects. However, under the special geomorphological background such as the high and cold mountainous areas in the northwest, it is difficult to obtain control points, which results in the difficulty of geometric correction by conventional methods.

The current GNSS is developing rapidly. The trajectory data of the car navigation system is more and more abundant. What's more ,The multi-spectral data of GF-1 images have richer spectral information and the full-color image has a higher resolution, thus to solve the problem without control points or control point image geometric correction under the condition of scarce problem, this paper adopts an image geometry correction method based on line features for GF-1 Nanjing urban image. Firstly, the center line of linear objects in the Remote sensing image is extracted by remote sensing technology, and the GNSS trajectories of these linear objects are measured. Secondly, the line features of the image and the measured GPS trajectory are constructed by the line feature description, and the corresponding line is obtained based on the line feature matching algorithm. Line pairs, and based on the homonymous line to establish a correction model, and then use a small number of identified control points of the homonymous line on the model to be optimized; finally based on the correction model of the image of the geometric correction, and the accuracy of image correction to make an evaluation. This article has obtained the following conclusions through research and analysis:

(1) Based on the C4.5 algorithm, the degree of intelligence and automation are improved in the process of road center line extraction. The experimental results show that the accuracy of the proposed method is 90% for the simple road extraction results, and the accuracy of the more complex road network extraction results is 87%, medium error is, which meets the accuracy requirement of the 2m resolution GF-1 Nanjing image.

(2) The homonymous line matching can be performed by describing the spatial structure and geometric relationship between the linear objects in the image and the measured GPS trajectory, and the homonymous line can be obtained quickly and accurately. The experimental results show that for the Nanjing GF-1 image, when the distance between the homonymous line pair is 31m, the rectangle width is 65m, and the slope ratio is 0.4, the matching accuracy is the highest.

(3) The correction model based on line features can solve the problem of image scale, rotation and translation, and then correct the polynomial model by a small amount of identifying the same point, which can improve the accuracy of correction. The final experimental results show that the medium error after correction is within, which confirms the feasibility of establishing correction models based on line features to correct the image in the absence of control points.

KEY WORDS: Information extraction, line feature matching, geometric correction, GF-1 image

目 录

摘 要 1

Abstract 2

第一章 绪论 1

1.1 研究背景及意义 1

1.2 国内外研究现状 1

1.2.1 遥感影像道路信息提取研究现状 1

1.2.2 特征线匹配研究现状 2

1.2.3 影像校正模型构建研究现状 3

1.2.4 存在问题与进一步研究目标 3

1.3 研究内容与技术路线 4

1.4 本章小结 5

第二章 实验数据与预处理 6

2.1 实验数据 6

2.2 影像预处理 7

2.2.1 辐射定标 7

2.2.2 正射校正 8

2.2.3 图像融合 9

2.2.4 大气校正 10

2.3 本章小结 11

第三章 道路中心线提取 12

3.1 C4.5算法概述 12

3.1.1 算法实现过程 12

3.1.2 信息增益率的计算 13

3.2 基于C4.5算法的道路中心线提取 14

3.2.1 建立特征空间 14

3.2.2 选择训练样本 16

3.2.3 生成决策树 17

3.3 分类后处理 18

3.4 实验结果与精度评价 20

3.4.1 C4.5法道路分类结果 20

3.4.2 道路中心线提取结果 22

3.4.3 精度评价 23

3.5 本章小结 25

第四章 同名线匹配 26

4.1 同名线对的特征描述 26

剩余内容已隐藏，请支付后下载全文，论文总字数：82493字