论文总字数：40784字

摘 要

合成孔径雷达是一种主动微波传感器，有着全天时、全天候的观测能力，可实现对地物目标的高分辨率成像，已在各个领域得到了广泛应用，是遥感对地观测的重要手段之一。基于高分辨率SAR影像进行场景与目标的三维重建，是当前SAR领域研究的前沿热点方向，在军事侦察、灾害监测、智慧城市建设等领域有重要应用价值。基于SAR影像的建筑物三维重建，因其在数字城市建设等方面的应用潜力，而得到了普遍关注。建筑目标的三维重建可获得不同地区的建筑分布和高度等信息，有助于城市数据的分析，可以为城市的发展方向和发展策略提供重要依据，具有重要意义。

本文开展基于高分辨率SAR遥感影像的建筑物三维重建方法研究，采用2016年苏州地区的Terrasar-X数据，以简单场景中建筑目标为研究对象，通过对建筑目标散射特征的提取和几何参数的计算实现了SAR图像建筑目标的三维重建，主要研究工作如下：

（1）SAR系统建筑目标成像机理及主要散射特征研究。

学习和研究了SAR系统的工作原理和建筑物的成像原理，分析了建筑目标的典型散射特征，总结了SAR图像中建筑物的特点和SAR建筑目标三维重建中特有的优势。

（2）SAR图像建筑目标检测研究。

基于霍夫变换与Otsu算法，提出基于自适应阈值区域分割和霍夫变换边缘检测的SAR图像建筑目标检测算法，实现了SAR图像中二次散射与叠掩区域的自动检测与提取。

基于马尔科夫随机场模型及相关理论基础，实现了基于MRF模型的检测算法。通过基于ICM算法的MRF模型，采用K均值算法进行聚类，GMM模型估计概率模型参数，对图像进行分类，并针对分类后图像存在强回波噪声干扰的情况，采用区域生长法，结合先验知识，提取建筑叠掩（或阴影）区域。通过不同场景中建筑物的检测证明了算法的有效性与适用性。对比两种算法，证明了基于MRF模型的检测方法更佳。

（3）建筑目标几何参数估计与重建

实现了基于立体几何反演的建筑目标参数估计方法，通过地距图上获取的数据进行验证，精度达到了95%，证明了算法的有效性。在此基础上，利用计算出的几何参数与提取的SAR图像建筑物散射信息进行建筑目标的三维重建。

关键词： 合成孔径雷达，建筑目标检测，Hough变换，MRF模型，三维重建

Abstract

Synthetic Aperture Radar (SAR) is an active microwave sensor with all-day and all-weather observation capability. As an important means for remote sensing of ground observations, it can achieve high-resolution imaging of ground objects and has been widely used in various fields.The three-dimensional reconstruction of scenes and targets based on high-resolution SAR images is a hot topic in the field of SAR research. It has important application value in military reconnaissance, disaster monitoring, and smart city construction. The three-dimensional reconstruction of buildings based on SAR images has received widespread attention due to its application potential in digital city construction. The three-dimensional reconstruction of the building target can obtain building distribution and altitude information in different regions, which can be help in analyzing urban data, providing advice for city development direction and strategy and is of great significance.

Our data source is the Terrasar-X data about Suzhou derived in 2016 and study targets in a simple scene. The 3D reconstruction of the SAR image building target was achieved through the extraction of scattering features and the calculation of geometric parameters. The main work is as follows:

(1)Research on imaging mechanism and main scattering features of SAR architectural targets.

Study the working and imaging mechanism of SAR. Analyze typical scattering features such as overlay, shadow, and secondary scattering of architectural objects in detail. Summarize the characteristics of buildings in SAR image and unique advantages in 3D reconstruction of SAR system.

(2) Research on building detection of SAR image.

Introduce the principles of Hough transform and Otsu algorithm, and basing on region segmentation and edge detection, propose an algorithm to detect and extract the scattering characteristics.

Basing on the Markov random field model and related theories, implement an image classification algorithm. Through the MRF model based on the ICM algorithm, the K-means algorithm is used for clustering, the GMM model estimates the parameters of the probabilistic model, and then image is classified. For the case where there is strong echo noise interference after the classification, the region growing method is used in combination with the prior knowledge, extracting architectural overlays (or shadows). The single and multiple architectural targets with distinct scattering features in simple scenes were classified and tested, and both obtained satisfactory results. The effectiveness and applicability of the algorithm is verified. The comparison between the two algorithms shows that the method based on the MRF model is better.

(3)Estimate geometric parameters and reconstruct the building.

A parameter estimation method for building targets based on three-dimensional geometry inversion was proposed. The accuracy was verified by the data obtained from the ground distance map, proving the effectiveness of the method. On this basis, the 3D reconstruction of the building target is performed.

KEY WORDS: SAR, building target detection, Hough transform, MRF model, 3D reconstruction

目录

摘要 I

Abstract II

第一章 绪论 1

1.1 选题背景及意义 1

1.2 国内外研究现状 1

1.2.1 SAR发展概述 1

1.2.2 高分辨率SAR图像建筑物检测研究现状 2

1.2.3 高分辨率SAR图像三维重建方法研究现状 5

1.3 本文的研究内容 6

1.4 研究难点 7

1.5 本文的结构安排 7

第二章 SAR系统原理及传统图像检测算法 9

2.1 SAR原理概述 9

2.1.1 SAR系统工作原理 9

2.1.2 SAR建筑物成像特点 9

2.1.3 SAR图像建筑物特点 11

2.1.4 SAR在建筑目标三维重建中的特有优势 11

2.2 传统SAR图像检测方法 12

2.2.1 Hough变换直线检测 12

2.2.2 Otsu区域分割算法 13

2.2.3 检测过程 14

2.3 本章小结 18

第三章 基于机器学习的SAR图像检测算法 20

3.1 MRF基础理论介绍 20

3.1.1 MRF模型基础 20

3.1.2 吉布斯（Gibbs）分布 21

3.1.3 MRF图像处理中的应用 22

3.2 基于MRF模型的SAR图像检测算法实现 26

3.3 实验结果 27

3.3.1 分类结果 27

3.3.2 分类后的处理 27

3.3.3 其他SAR图像建筑目标检测 28

3.4 两种算法的比较 30

3.5 本章小结 30

第四章 SAR建筑目标的三维重建 32

4.1 建筑目标的几何参数估计 32

4.1.1 建筑物高度计算方法 32

4.1.2 建筑物长宽尺寸计算方法 32

剩余内容已隐藏，请支付后下载全文，论文总字数：40784字