MAPSAT TELERILEVAMNETO MEDITERRANEA (IT)
The project is a follow up of the ESA funded OSIRIS project, aimed at developing a series of modules for maritime surveillance into a prototype deployed on a shared processing system with a tested full chain for Sentinel 1 and 2 data. The team performed a survey on most commonly used spaceborne sensors for thermal infrared, selecting four sensors for the demonstration phase, characterized by better values of resolution and accuracy and different equatorial crossing times.
The team set up the full chain of processing for Temperature maps extrapolation, based on processing of Top of the Atmosphere (ToA) radiance, brightness temperature at the sensor, and land masking. An impact assessment study was conducted the issue of calculating and using of the Sea Surface Temperature map instead of the atmospherically correct products. The interface of the Ship Detection module based on the CFAR detection algorithm is being updated for the ingestion of temperature map raster. As for the Ship Classification module, the ground-truth database has been finalized. The feature extraction code was modified to automatize the access to OpenSARShip ( open dataset), and to provide the updated features, which are being reconsidered during the study of the classification algorithms. The team is working to improve the ship classification methods from SAR HR images, focusing on Random Forest algorithm, with average accuracies of about 70% for noisy data and 80% for ground-truth data, on four ship types, compared to other methods (Naive Bayes and Multiple Kernel Learning).
The kinetic parameters extraction module exploits single look complex (SLC) SAR images and a clutterlock algorithm to estimate the doppler frequency shift. The team is also advancing Ship Behaviour Analysis based on Sentinel 2 and sat- AIS historical datasets from AstraPaging and ExactEarth.