Smarter, smaller AI models outperform bigger rivals at spotting mussel

A study focused on mussel farming off the coast of Galicia, Spain, suggests that smaller AI models can outperform larger ones for detecting mussel rafts from satellite images. Researchers published in Frontiers in Remote Sensing used very high-resolution imagery over the Ría de Arousa, a hub where nearly 70% of platforms across Galicia’s Rías Baixas are concentrated. Spain supplies about 44.6% of Europe’s mussel output and roughly 10% of global production. The work compares YOLO detectors that produce axis-aligned boxes with oriented bounding box (OBB) versions designed for rotated, elongated rafts (“bateas”). Oriented boxes aim to reduce center and size errors that can affect concession compliance, environmental checks and navigation.