Automatic biomass estimation of European sea bass using machine learning and computer vision technique

In recent years, the aquaculture sector has been rapidly developing to meet the increasing global demand for protein, and this progress has created a growing need for more efficient and sustainable production methods. In particular, accurately, quickly, and non-invasively measuring the biomass of fish is of great importance for optimizing feeding strategies, determining harvest timing, and monitoring overall stock health. However, traditional biomass estimation methods rely on manual sampling and weighing processes, which lead to time loss and increased labor costs. Moreover, removing fish from the water can elevate stress levels, damage the mucus layer, and negatively affect animal welfare. For all these reasons, the need for contactless, fast, and highly accurate biomass estimation systems in the aquaculture industry is increasing day by day. In this thesis study, an innovative system was developed for automatic biomass estimation using image processing and machine learning techniques, with a particular focus on the European sea bass (Dicentrarchus labrax). Within the scope of the study, different versions of the YOLO architecture YOLOv5, YOLOv7, YOLOv8, YOLOv9, and YOLOv11 were tested, and their segmentation performances were comparatively evaluated. As a result of the experimental analyses, YOLOv8 was identified as the most accurate model and was therefore integrated into the segmentation component of the system. The developed system processes 15,000 underwater images obtained from fish farms, automatically extracting the length and width measurements of the fish through segmentation. The extracted morphological features were used as input for regression algorithms to perform biomass estimation. Among these, the XGBoost algorithm achieved an R² value of 0.981 and emerged as the most successful model. Operating on portable hardware, the system enables contactless measurement, thereby reducing stress and injury risks in fish, and providing ease of use in field conditions. The results obtained demonstrate that this technology offers a sustainable and innovative solution for fish farming.