高分辨率合成孔径雷达图像的直线特征多尺度提取方法

Multi-scale extraction approach of linear feature in high resolution SAR images

针对传统的合成孔径雷达(SAR)多尺度边缘提取方法中直线提取连续性和完整性不好的特点,提出了一个由粗到精的多分辨率SAR图像直线特征多级提取框架,利用多尺度策略在降低SAR图像噪声影响的同时增强相邻共线点之间的连续性和完整性,并在不同的尺度图像中根据边缘特征的特点选择不同处理方法,来实现低分辨率条件下完整直线特征的粗略提取和高分辨率的精确定位。最后用高分辨率SAR图像跑道检测实验进行了验证,并将实验结果与相位编组法和Hough变换法进行了比较。

In this paper,a multi-level frame was designed to extract linear feature in multi-resolution Synthetic Aperture Radar（SAR）images,to improve the extracted line＇s continuity and integrity by traditional methods.The speckle filtering and the linear feature extracting were executed synchronously and parallel,the different edge detection and edge points grouping method were used to extract linear feature coarse-to-fine according to the characteristics of collinear points in different scale images,and the linear features were extracted continuously,completely,accurately and efficiently.At last,the approach was tested by the runway extraction and road extraction;the results were analyzed with phase group and Hough transform.