Low Power Naval surveillance Radar for low RES Targets in Rough Seas – NEPTUNE 3

BOWEN has developed a NEPTUNE program that aims to have innovative algorithms to ensure processing that guarantee the detection and classification of small objects (SER of less than 3 m2) by rough seas (up to level 5 with a SCR of less than 10dB), this with a low power onboard pulse radar (10-40W).

This program resulted in a RAPID (NEPTUNE 1 from 2016 to 2019) in collaboration with ONERA and two CIFRE Defense theses (NEPTUNE 2 from 2019 and 2023) in collaboration with the University of La Rochelle. This latest work has led to the development of signal processing based on imaging analysis logic from the medical field. At the same time, BOWEN has worked on the radar sensor, innovative processing based on the use of parsimonious decompositions as well as more conventional processing. These processing provide information for characterization of the environment such as sea states, coastal features or rain zones. This work showed that the introduction of a global supervision of the processing based on a better knowledge of the marine context could optimize by learning the parameterization of the algorithms, as well as select the treatment providing the best result.
The main purpose of the next block of the program (NEPTUNE 3) is therefore to set up a Meta-Learning platform ensuring that the methodology used will be the most effective, considering the weather and sea conditions encountered and ensuring its rapid adaptation to a change of environment. NEPTUNE 3, proposed in collaboration between BOWEN, ENSTA Bretagne, L2S and SONDRA, therefore requires on the one hand the optimization of various developed processing algorithms and on the other hand the establishment of a learning database (metadata) results from various signal processing applied to I&Q data from the NEPTUNE 2 sensor or equivalent sensors. The laboratories of ENSTA Bretagne (Lab-STICC), SONDRA and L2S will contribute their know-how to this process.

The NEPTUNE 3 project thus proposes to make NEPTUNE 2 processing more robust, by optimizing them according to the perceived sea state, and to complete them by taking into account the signal phase in the processing, as well as algorithms based on information geometry analysis. In this context dedicated to data and related processing, the objective is to feed the metadata database with contextual environmental data and richer detection/classification data. ENSTA Bretagne (Lab-STICC), SONDRA and BOWEN will work on these objectives.

Many of these processes use machine learning processes, so it is important to also enrich the existing data to make them more robust. The project uses an existing database, but on the one hand will enrich it with real data collected with the sensor developed in NEPTUNE 2 and the HYDRIX radar (Mont Vial BOWEN weather radar), but also by VAE and GAN enrichment processes designed to broaden the scenarios considered and by the addition of simulated targets difficult to obtain in real scenarios such as periscopes or floating plates. L2S and SONDRA will work mainly on the first aspects, ENSTA Bretagne (Lab-STICC) on models and simulations to enrich the database with cases.

Finally, it is necessary to be able to industrialize these solutions and thus to integrate the algorithms selected in an on-board computer. In the project, BOWEN will take over the integration aspects of the selected solutions and the constraints induced on the embedded algorithms, as well as the data architecture.