Extraction method of large-scale linear objects from SAR images

Public safety concerns have been raised by potential safety problems due to deformation of infrastructures such as railways, highways and bridges. Monitoring the deformation of large-scale man-made linear objects (LMLO) by permanent scatters (PS) interferomerty would be critical to prevent this kind of disasters. However, the estimation of LMLO deformation is sensitive to the influence of those random PS from nearby natural objects. Thus, extraction of LMLO from PS images is the necessary condition for an accurate monitoring of LMLO deformation. In this paper, a new procedure of extraction LMLO from SAR images is proposed based on classification, morphological operations and line recognition methods. A preliminary step, merging optical image, would be identification of man-made objects regions from background. Image of permanent scatters (PS) points from PALSAR data will then be masked to remove unrelated PS points. Then a set of morphological operations and line feature extraction methods are performed to connect PS points continuously distributed along the linear target. This method is successfully applied to a PALSAR image to extract the Lantau Link highway (LLH), Hong Kong and the experiment result demonstrates its validity on comparison with the ground truth in optical image.