SAR data fusion and a novel joint use of neural networks for coastline extraction

ABSTRACT

In this paper, a novel automated coastline extraction method from SAR (Synthetic Aperture Radar) data is presented. The method is designed to exploit radar backscatter coefficients (${\sigma }^0$) from multipolarization SAR acquisitions (the 4 classic co- and cross-polarized polarizations), whereas single-pol data are employed in the majority of methods in this field, implementing data fusion through the use of an autoencoder neural network and producing the coastline by harnessing a Pulse-Coupled Neural Network (PCNN). Main results are presented throughout the paper, demonstrating superiority and comparability with established methods and with a recent automated algorithm that can be considered among the state-of-the art techniques in this field; furthermore, effectiveness of data fusion and segmentation obtained through the mentioned neural networks has been compared to that of several combinations of the same networks with different frameworks: a different data fusion framework, obtained through the use of linear Principal Component Analysis (PCA), and a different binarization framework, based on the use of Expectation-Maximization (EM) image segmentation. Main achievements of presented technique consist in enabling a possible faster processing as well as the opportunity of operating with an improved fused information content on coastline, together with very high accuracy results.

Acknowledgements

Part of this research has been carried out within the framework of SOAR-ASI project n${ }^o$ 5227. RADARSAT-2 data are provided by Canadian Space Agency: RADARSAT-2 Data and Products $\copyright$ MacDonald, Dettwiler and Associates Ltd. (2010) - All Rights Reserved. RADARSAT is an official trademark of the Canadian Space Agency. The authors give thanks to USGS for providing freely available Landsat data and to Google for the free online supply of maps and satellite scenes. Landsat data are available under a free and open data policy at https://earthexplorer.usgs.gov/. RADARSAT-2 data are available at https://www.eodms-sgdot.nrcan-rncan.gc.ca/index-en.html.

Disclosure statement

No potential conflict of interest was reported by the author(s).