Coastline monitoring over time is crucial to promptly detect and address environmental problems such as coastal erosion. Satellite imaging offers a great opportunity for this kind of tasks, but proper analysis tools are required to identify sea and land regions. Several techniques have been proposed over time for satellite images analysis, typically based on the direct computation of a water probability index for each pixel. In more recent years, however, research was focused on the usage of deep learning techniques for sea-land segmentation and coastline detection. For these methods, a large dataset of labelled samples is required but often not available. In this paper, we propose a method for the automatic generation of a dataset of labelled satellite images, containing both sea and land regions. The automatic labelling method is based on the combination of information retrieved from publicly available coastline data and from satellite images themselves and can be used to generate a large number of sea-land segmented samples.
A new dataset of satellite images for deep learning-based coastline measurement / Scarpetta, Marco; Spadavecchia, Maurizio; D'Alessandro, Vito Ivano; Palma, Luisa De; Giaquinto, Nicola. - ELETTRONICO. - (2022), pp. 635-640. (Intervento presentato al convegno 2022 IEEE International Conference on Metrology for Extended Reality, Artificial Intelligence and Neural Engineering (MetroXRAINE) tenutosi a Rome, Italy nel 26-28 October 2022) [10.1109/MetroXRAINE54828.2022.9967574].
A new dataset of satellite images for deep learning-based coastline measurement
Scarpetta, Marco;Spadavecchia, Maurizio;D'Alessandro, Vito Ivano;Palma, Luisa De;Giaquinto, Nicola
2022-01-01
Abstract
Coastline monitoring over time is crucial to promptly detect and address environmental problems such as coastal erosion. Satellite imaging offers a great opportunity for this kind of tasks, but proper analysis tools are required to identify sea and land regions. Several techniques have been proposed over time for satellite images analysis, typically based on the direct computation of a water probability index for each pixel. In more recent years, however, research was focused on the usage of deep learning techniques for sea-land segmentation and coastline detection. For these methods, a large dataset of labelled samples is required but often not available. In this paper, we propose a method for the automatic generation of a dataset of labelled satellite images, containing both sea and land regions. The automatic labelling method is based on the combination of information retrieved from publicly available coastline data and from satellite images themselves and can be used to generate a large number of sea-land segmented samples.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.