Coastline Extraction in Remotely Sensed Images

Ferdinando Giordano, Dellepiane, Raimondo De Laurentiis

Dept. of Biophysical and Electronic Engineering - University of Genova (IT)

The coastal zone is a vital, and highly dynamic environment. Within such area, multiple geophysical parameters are important to monitor in order to understand in deep the evolution of it. Depending on the objectives targeted and on the particular phenomenon to monitor, different resolutions, both on a temporal and on a spatial scale, are required. This is due to the fact that each phenomena, as erosion or accretion of the coastline or variations in typology or volume of sediments, has different time scale evolution. Thus, when considering the exploitation of remote sensing techniques for monitoring purposes, a choice has to be made with respect to the best satellite data to be chosen.
The dataset used in the present work for the extraction of the coastline is composed by SAR images acquired by ERS-1 and ERS-2 satellites.
Such images are characterised by a spatial resolution of 25x25 meters. The choice of exploiting data with such a low resolution can be seen as a drawback, considering the fact that the objective of the research activities is to provide a valuable method for the extraction of coastline for monitoring purposes.
Moreover, traditional methodologies provide a much higher resolution with respect to the one obtained from SAR data. Nevertheless, the activities have been focused on SAR images for the fact that, in the next future, satellites with a few meter resolution SAR sensors will be launched. In particular COSMO SkyMed will provide high-resolution SAR data for marine applications especially for the Mediterranean Sea. As already mentioned, SAR data are attractive as deals with the possibility of acquisition regardless weather conditions. This facility is particularly important for application such as the monitoring of coastline, as most relevant changes are related to bad weather conditions and sea storms.
The methodology here proposed for the extraction of the coastline is based on the analysis of the remote sensing SAR data, taking polarimetric and multitemporal aspects into account at a same time. In fact, image intensities and the properties of interferometric coherence derived from the correlation of the InSAR couple are exploited together.
The proposed approach is a semi-interactive segmentation based on a seed- growing process. The growth starts from the seed point indicated by the user as surely belonging to the object searched for. Then it follows the best paths in terms of connectivity, thus guaranteeing the extraction of a connected structure and of its fine details.
Due to the fuzzy nature of the algorithm, the final result is not hard but fuzzy. The final image, named the connectedness map is characterised by membership values that indicate to what extent every pixel is connected with the seed point. This means that the user can easily choose his best result by thresholding the connectedness image.
The proposed algorithm which is applicable to other type of data (for example optic images) is mainly data-driven, and makes use of a few information it can extract from the indicated seed point, as characteristic of the desired region. It could therefore be considered as partially supervised, having a few indications about the searched class or used for other.