高分辨率SAR图像城区道路半自动提取方法研究

As the typical element of geographic features, road has been widely used in various fields of administration such as urban planning, GIS database updating and so on. Road extraction from SAR images is a basis and forefront research topic of geographic features interpretation. Due to obstacles blocking, shadowing and layover effects, it is difficult to extract roads from urban high resolution SAR images by automatic methods. These issues such as time-consuming, poor robustness and high false alarm rate still exist. Semi-automatic methods can combine computer's fast calculation with human's interpretation skills, which is still a promising research area. This project takes the semi-automatic road extraction framework as reference. Firstly, road junction detection and extraction is investigated based on its radiation and geometry characters, which could provide initial points and orientation for tracking. Then, double window model is established to detect road local information like orientation, width and position. The local main orientation is calculated in outer window, which combines the methods of CFAR road region segmentation with parallel detection. The local road width and position of center point are calculated in inner window. Finally, road center point tracking is carried out by investigating ways which integrate local contexts with particle filtering. Through this project, the degree of applicability and veracity of road extraction would be effectively improved.

道路作为典型的地物要素，在城市规划、GIS数据库更新等领域应用广泛。SAR图像道路提取是地物要素解译的基础前沿课题。由于城区道路周边障碍物遮挡、建筑物阴影等干扰严重，实现高分辨率SAR图像城区道路自动提取非常困难，现有算法存在计算时间偏长、适应性不强、虚警率偏高等问题。半自动提取方法能有效结合机器的快速计算和人的解译技巧，在当前仍不失为一种较有前途的研究方向。 本项目参照半自动道路提取架构，先根据道路交叉口的辐射及几何特征，研究交叉口检测和识别方法，为后续跟踪提供初始点和前进方向。然后，建立双窗口局部检测模型，在外窗口内研究CFAR道路分割及平行线对检测方法，计算道路局部主方向；使用内窗口计算道路的宽度和中心点坐标。最后，研究结合局部上下文信息的道路中心点粒子滤波抗干扰跟踪方法，实现高分辨率SAR图像城区道路快速有效提取。通过本项目研究，有望提高道路提取方法的普适性和准确性。

道路在城市规划、GIS数据库更新等领域应用广泛。该项目瞄准当前热点的高分辨率SAR图像道路要素解译问题，研究实现了3个方面内容。①根据道路交叉口的辐射、几何等特征，通过计算Zernike矩特征，采用支持矢量机的方法实现了道路交叉口的有效提取；②采用Lee算法滤波、Canny边缘检测和Radon变换等三步骤法实现了道路主方向及宽度的计算；③采用粒子初始化、粒子传播、权值更新、粒子估计及中断处理等五步骤法实现了道路中心点的粒子滤波跟踪。研究方法克服了城区道路环境背景复杂、存在各类干扰以及与其他地物对比度不明显等带来的道路提取难题，总体提高了城区道路中心点提取的准确率和降低虚警率，在减少人机交互次数、提高自动化程度的同时，有效的实现高分辨率SAR图像城区道路提取。项目延续将结合用户需求进一步拓展军民应用。