Automatic detection of fin, operculum and skin deformities in Mediterranean Fish Species

Morphological deformities in farmed fish can significantly impair swimming performance, feeding efficiency, and overall welfare, leading to economic losses in aquaculture. In this study, we apply deep learning-based computer vision models to automatically detect pectoral and caudal fin deformities as well as skin and operculum deformities ("red spots") in two key Mediterranean species: European seabass ( Dicentrarchus labrax ) and gilthead seabream ( Sparus aurata ). Using underwater stereo-vision imagery from our pilot-scale fish farm, we developed diverse annotated datasets for each morphological feature through a combination of manual labeling, an AI-assisted annotation tool and data expansion pipelines (data augmentation). We trained and evaluated multiple computer vision model architectures to address underwater imaging challenges such as motion blur, occlusion, and variable light intensities. Across all model comparisons, the lightweight YOLOv12s model achieved the best balance of accuracy and computational efficiency, with mean Average Precision values up to 0.81 for fin classification and 0.91 for red spot detection, while enabling real-time inference. Our results demonstrate that small CNN models, improved with attention-based methods like the YOLOv12s and trained on diverse, well-prepared images, can assess fish morphology in real aquaculture environments. This approach offers a scalable, non-invasive monitoring solution to support real-time monitoring and early health interventions and improve fish welfare in intensive farming systems. • We present a tool that detects fin, operculum and skin deformities in Mediterranean fish. • We collected image data for annotation at our pilot farm and trained multiple AI models. • YOLOv12s was the best performing model combining both, high precision and speed. • This lightweight model can enable real time detection applications in Mediterranean aquaculture. • Our work supports an enhanced fish welfare and improved farm health management and performance.