AI-based automated monitoring of the invasive pearl oyster (Pinctada radiata) in the Aegean Sea using underwater surveys

The Mediterranean Sea is increasingly affected by invasive alien species due to increased shipping traffic through the Suez Canal and ongoing ocean warming. The pearl oyster Pinctada radiata (Leach, 1814) is one of the most widespread Lessepsian invaders (i.e., species that have migrated from the Red Sea to the Mediterranean Sea through the Suez Canal), yet standardized, non-destructive monitoring approaches for this sessile bivalve remain limited. Here, we present a proof-of-concept AI-based framework for automating the monitoring of P. radiata using diver-operated in situ underwater video surveys conducted in the Southern Evoikos Gulf, Greece. The approach integrates a YOLOv11 detection model with the ByteTrack tracking algorithm and incorporates an image-based morphometric module to estimate oyster size and weight. The detection model achieved promising performance across heterogeneous benthic habitats (F1 score = 0.85; mean average precision (mAP) = 0.845). Automated abundance estimates closely matched manual counts, with 53 oysters detected compared to 51 manually identified ground-truth individuals. Video-derived shell-length distributions (30–70 mm) were consistent with independent field observations from the study area. This framework demonstrates the potential of AI-assisted underwater video analysis for extracting abundance and population structure metrics of invasive bivalves. However, it was developed and evaluated within a single study area with limited data coverage, weight estimation depends on a pre-existing empirical length–weight relationship derived from destructive sampling, and individual-level morphometric accuracy remains unvalidated. Future work incorporating expanded datasets, direct individual-level validation, and diverse Mediterranean Sea coastal ecosystems is required to evaluate broader applicability.