DEMO: Automated fish counting & species recognition with ai video analysis
Monitoring fish assemblages in coastal waters is essential for assessing biodiversity and ecosystem dynamics, particularly under increasing anthropogenic pressures and climate change. However, traditional approaches, including visual censuses and net sampling, are labour-intensive, costly, often invasive, and provide limited temporal resolution. To address these limitations, the Ruđer Bošković Institute (RBI) established a permanent underwater camera system ten years ago at approximately 5 m depth in front of its marine research station, located within the Natura 2000 site “Krka Mouth” near Šibenik, Croatia. Over this period, the system has generated a substantial archive of underwater photographs and selected short video recordings. Initially, the material was analysed manually, demonstrating the value of continuous visual observations but also highlighting the need for automated processing. The system was subsequently upgraded with an improved underwater camera equipped with an automatic lens-cleaning mechanism. In collaboration with BluedataB, RBI developed and trained a machine-learning model for the automated detection, classification and tracking of Adriatic fish species. The model currently recognizes 46 species, tracks individual fish within the camera field of view and automatically counts recorded individuals. Combining AI-based image analysis with long-term underwater observations provides a non-invasive, high-temporal-resolution approach to fish biodiversity monitoring. Importantly, automated analysis also creates opportunities to exploit the existing ten-year image archive for assessing long-term changes in coastal fish assemblages and supporting the management of marine protected areas.