Two-step algorithm is proposed for unsupervised detection of linear structure from SAR image, in particular, the road network detection. The first step is to detect line element with local detector, in this paper an improved ratio of average (RoA) detector which takes into account the inner statistic property of SAR image is proposed. It can guarantee the detection accuracy. In the second step, we introduce the prior knowledge about the road structure to group the line element into macroscopical structure. The method proposed is identified in ERS-1 image.%提出一种两步算法用于从合成孔径雷达(SAR)图像中无监督地提取线性特征,特别是提取公路网。第一步利用局部线特征检测算子提取线基元,为此提出一种改进的均值比率(RoA)算子,应用SAR图像的内在统计特性,保证较高的准确率;第二步引入公路结构特征的先验信息,对提取的线基元进行组织,获得比较宏观的线性结构。文中的方法通过ERS-1图像加以验证。
展开▼