Maritime traffic monitoring is a long-standing research topic in remote sensing, with major implications for security, environmental protection, and surveillance of illegal activities. Phenomena such as unreported fishing, smuggling, human trafficking, and the deliberate sinking of unregistered vessels demonstrate the urgent demand for effective tools capable of continuous and large-scale observation of critical areas such as the Mediterranean basin. Current Earth observation missions provide valuable data for ship detection, yet relying on a single mission limits the achievable temporal and spatial coverage. The SPECTRE project (ShiP dEteCTion based on aRtificial intElligence; agreement n. 2023-4-E.0), funded by the Italian Space Agency under the RESEARCH DAY ASI 2020 program, addresses this challenge through the development of an artificial intelligence framework for the integration of multimission satellite data. The system leverages both radar and optical constellations, including Sentinel-1/2, COSMO-SkyMed, and upcoming missions such as IRIDE, to increase revisit times and enable vessel tracking over wide maritime regions. Deep learning techniques are employed throughout the processing chain: super-resolution models improve the interpretability of optical images, while tailored detection algorithms are trained on SAR data. A key component of the project is the construction and characterization of a dedicated dataset based on COSMO-SkyMed acquisitions. Images in different polarizations and resolutions are cropped and annotated using automated workflows, with vessel positions matched against interpolated AIS tracks. The resulting dataset provides standardized crops, polygonal annotations, and metadata suitable for training state-of-the-art detection networks. Finally, the SPECTRE system integrates detection outputs with a spatio-temporal matching algorithm to correlate multi-mission observations and track vessels across acquisitions. The project foresees the deployment of a WEBGIS platform to make these capabilities available for the operational monitoring of non-cooperative ships.
An AIS-Integrated COSMO-SkyMed Image Dataset For Ship Detection / Graziano, M.D., Mazzeo, A., Cristofano, A.C., Galario, A., Di Vaia, M., Renga, A., Pisacane, V., Aurigemma, R., Focone, M., Lo Russo, A., Ravellino, M., Volpe, A., Tapete, D., Virelli, M.. - (2025), pp. 482-486. (2025 IAF Earth Observation Symposium at the 76th International Astronautical Congress, IAC 2025 aus 2025) [10.52202/083081-0058].
An AIS-Integrated COSMO-SkyMed Image Dataset For Ship Detection
Graziano M. D.;Mazzeo A.;Cristofano A. C.;Galario A.;Renga A.;
2025
Abstract
Maritime traffic monitoring is a long-standing research topic in remote sensing, with major implications for security, environmental protection, and surveillance of illegal activities. Phenomena such as unreported fishing, smuggling, human trafficking, and the deliberate sinking of unregistered vessels demonstrate the urgent demand for effective tools capable of continuous and large-scale observation of critical areas such as the Mediterranean basin. Current Earth observation missions provide valuable data for ship detection, yet relying on a single mission limits the achievable temporal and spatial coverage. The SPECTRE project (ShiP dEteCTion based on aRtificial intElligence; agreement n. 2023-4-E.0), funded by the Italian Space Agency under the RESEARCH DAY ASI 2020 program, addresses this challenge through the development of an artificial intelligence framework for the integration of multimission satellite data. The system leverages both radar and optical constellations, including Sentinel-1/2, COSMO-SkyMed, and upcoming missions such as IRIDE, to increase revisit times and enable vessel tracking over wide maritime regions. Deep learning techniques are employed throughout the processing chain: super-resolution models improve the interpretability of optical images, while tailored detection algorithms are trained on SAR data. A key component of the project is the construction and characterization of a dedicated dataset based on COSMO-SkyMed acquisitions. Images in different polarizations and resolutions are cropped and annotated using automated workflows, with vessel positions matched against interpolated AIS tracks. The resulting dataset provides standardized crops, polygonal annotations, and metadata suitable for training state-of-the-art detection networks. Finally, the SPECTRE system integrates detection outputs with a spatio-temporal matching algorithm to correlate multi-mission observations and track vessels across acquisitions. The project foresees the deployment of a WEBGIS platform to make these capabilities available for the operational monitoring of non-cooperative ships.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
