AI for marine vessel detection

The application runs onboard Φsat-2 and makes use of machine learning techniques, in particular deep learning, to autonomously detect and classify vessels based on the images acquired through the Φsat-2 multispectral camera.  

The application inspects the image to determine some key features such as the absence of clouds, presence of ocean and the presence of ships in the image and then establishes whether or not a given scene (or area) requires further monitoring. Based on this, the sensor could further acquire data and transmit the data to the ground for further analyses.

This will enable faster responses, and potentially higher value detections. The autonomous vessel awareness application specifically targets: 

• an improvement in the quality of downlinked imagery for the purposes of vessel detection 
• a reduction in the volume of downlinked images 
• a reduction in time spent processing and analysing images by human operators 
• a reduction in the cost of operating satellite missions 

It is expected that the application will be able to detect ships longer than 40 m, create a patch of the full image that includes the specific ship, estimate its position and send the information to the ground.  

The information provided by the Φsat-2 mission can then be correlated with Automatic Identification System transponders, which are designed to automatically provide position, identification and other information about vessels to other ships and to coastal authorities. 

AVA is developed by CEiiA in Portugal.