The exploitation of Sentinel-1 images for vessel size estimation

Abstract

In this article, a novel technique for fully automatic vessel size estimation using medium-to-high-resolution synthetic aperture radar (SAR) images is presented. Based on mathematical morphology, it aims at better delineating the vessel outline in the cluttered SAR image, thereby enabling the extraction of its actual dimensions. The technique has been tested on a set of 127 ships representing a range in lengths between 24 and 366 m in five Sentinel-1 images at 20 m multi-look resolution that have good quality ground truth available. It is found that the proposed algorithm produces very good length estimates (15% relative error/30 m absolute error) and reasonable width estimates (35%/11 m). The estimates are significantly better than those from a simpler automatic method that does not use mathematical morphology, and approach those from manual analysis by an expert.