BRE03 Dive, Scan, Discover: Real-Time Fish Recognition Underwater

We want to develop an underwater camera that identifies fish species in real time and turns every dive into a citizen-science contribution. Recognition is performed by a lightweight computer vision model embedded directly in the camera, since no connection to a server is possible underwater. 

Because the model runs locally on very constrained hardware, the camera can be used by any diver or snorkeler, anywhere, with no dependency on network infrastructure. Each identified encounter is logged with species, timestamp and GPS location. After the dive, users sync their logs to a companion app offering a gamified history of species encountered, and can opt in to share sightings with marine biodiversity databases used by researchers for fishery resources management, wildlife conservation, training other classification models, etc. 

The camera also has an educational use, for school snorkeling outings or tourist excursions, and can be used directly by marine biologists and conservation researchers in the field to get real-time species counts and automatically logged transects, without reviewing hours of footage afterward. 

The core innovation is the miniaturization of the model to unlock mobile, offline, participatory marine monitoring and education that anyone can take part in.

Over the 48 hours, we will focus on three things.

First, building a lightweight fish-species recognition model, optimizing it to run with the lowest possible memory footprint on embedded hardware, without sacrificing accuracy — adapting the whole pipeline to the constraints of miniaturized components.

Second, the embedded architecture that captures a photo underwater, runs inference on-device, and displays the result quickly, with a UX designed for underwater conditions.

Third, a mockup of the citizen-science companion app: logging dives, scans, timestamp and location, and previewing how a user's dive history and contributions would look.

By the end of the hackathon, we want a working demo: the camera identifies a fish offline, logs it with time and location, and syncs to an app showing both the personal dive history and the citizen-science contribution.

Programming, computer vision, data organization, user experience, business model, design