Page 42: of Marine Technology Magazine (November 2025)

SYRENE UNDERWATER AI CAMERA

Credit: IFREMER S. Barbot

Credit: IFREMER S. Barbot

Realtime embedded detection with SYRENE.

This extra power also allows for the integration of Trans-

Toward SYRENE V2 and Small Vision Lan- former-type networks [2] as well as small vision-language guage Models (sVLMs) models (sVLM) [3]. These models favor a textual description

Updated in 2025, SYRENE has bene? ted from the hard- ware and software advances of recent years. We turned to an of the image rather than a direct transfer of the image, thereby embedded GPU from Nvidia (Jetson nano Orin NX) to test achieving signi? cant information compression. This is partic- the potential of a system where power consumption would be ularly useful for acoustic links with seabed equipment, where less of a constraint. This time, the entire system consists of a detector such as SYRENE will be deployed.

Very soon, it will be possible to integrate autonomous em- two modules: a low-power AI module and an AI module with bedded AI agents designed to perform speci? c tasks, such as higher computing power.

Depending on the trigger conditions and remaining battery sending SMS alerts in the event of speci? c events (e.g. species counting). These agents will be able to analyze data locally, life, the low-power module switches the embedded GPU, pro- make decisions in real time and act independently, while lim- viding embedded processing power for denser data and there- iting the energy consumption of embedded systems.

fore more accurate detection than before.

Conclusion

Ultra Low Power AI

Evolving to keep pace with recent technological advances in

The module, designed for continuous operation, is based on a microcontroller incorporating an embedded convolutional AI, SYRENE is an example of an oceanographic application that improves environmental monitoring and anticipates the impact neural network (CNN) accelerator, optimized for very low en- ergy consumption. The embedded AI unit enables it to ? nely of invasive marine species, particularly lion? sh. Coupled with ? lter relevant events and trigger the high-computing power an antenna and an IoT (Internet of Things) modem for shallow- water applications such as lagoons, or via a surface relay buoy section only when necessary.

In the future, new types of more advanced neural networks, for benthic applications, this type of architecture opens up a wide range of possibilities. It enables long-term monitoring of parame- such as spiking neural networks (SNNs) and adapted proces- sors, could be tested for the purpose, further reducing energy ters such as acoustics, video, seismicity and other physicochemi- cal parameters, as well as the compression of useful information.

consumption while improving detection quality.

In conclusion, the SYRENE detector thus offers new per- spectives for real-time monitoring of underwater fauna via

AI with high computing power

This new module can now integrate more powerful AI models imaging, acoustics or other parameters.

such as YOLOv11, which signi? cantly improves detection per-

Acknowledgments formance. This system allows YOLOv11 to take full advantage of

We would like to thank Pierre Ternat et Dominique Bartele- the GPU's parallel computing capabilities, bene? ting in particu- my (Oceanopolis) for having kindly hosted the camera in their lar from the optimizations offered by TensorRT, thereby reducing lion? sh tank, Helene Leau (Ifremer), SincObs project coordi- real-time latency by a factor of 4 in some cases we have tested.

Re-training (? ne-tuning) the neural network signi? cantly nator as well as Amaury Tisseau, Thomas Morales and Alvaro improved detection and processing performance. According Scarramberg for their involvement in this project.

to our tests, the computing power/energy consumption ratio

Credits with the old hardware is greater than six, opening up new pos- [1] Green, S.J., et al. (2012) sibilities in terms of image resolution: this translates into a [2] Vaswani et al., 2017 : https://arxiv.org/abs/1706.03762) greater detection distance, allowing smaller or more distant [3] https://hal.science/hal-04889751 objects to be detected with greater accuracy. 42 November/December 2025

MTR #8 (34-47).indd 42 MTR #8 (34-47).indd 42 11/20/2025 5:52:40 PM11/20/2025 5:52:40 PM