SPACEKNOW, INC., odštěpný závod (CZ)
The SEA-SPARK project aims to design, implement and experimentally evaluate deep learning based advanced algorithms for high-resolution Synthetic Aperture Radar (SAR) satellite imagery to detect and classify maritime vessels ranging from small transport and commercial vessels to large tankers and military ships. SpaceKnow will contribute to the state-of-the-art by improving performance and creating a robust and trustworthy solution, delivering better insights to customers reliant on precise and persistent maritime monitoring, vessel detection, and classification.
The specific applications range from maritime and coastal law enforcement (illegal fishing, migration, embargoed goods transport, pollution monitoring, detection, and prevention), port and coastal security to complex geopolitical issues like maritime border surveillance, exclusive economic zone sovereignty, and national security.
Special focus will be on exploration, recognition, and analysis of hard examples for training of deep neural networks. Hard examples are usually defined by their unique and diverse attributes or appearance, to be found very sparsely, thus creating a significant imbalance with respect to the rest of the data. The results and comparisons will be demonstrated by thorough evaluations on manually labeled testing data selected specifically to benchmark the ability of the DNN to perform well on these hard datasets.
