Floating Cage Detection for Trout Aquaculture in Juli and Pomata Bays, Lake Titicaca (Peru), Based on Sentinel-2 Satellite Images

Aquaculture in high Andean lakes has expanded rapidly in recent decades, particularly through the installation of floating cages for the production of rainbow trout (Oncorhynchus mykiss). However, this growth has occurred with limited environmental monitoring, leading to potential ecological impacts such as eutrophication, dissolved oxygen depletion, and loss of biodiversity due to nutrient accumulation and organic waste discharge.This study evaluates the detection of floating cages in a high Andean lake using RGB bands from Sentinel-2 imagery (10 m spatial resolution) and YOLO-based models, which are widely recognized for their efficiency in real-time object detection tasks. Comparative experiments were conducted among different versions of YOLO, achieving the best performance with the YOLOv8m model, which reached an mAP50 of 0.5474 in training and 0.5453 in validation. The results were mainly influenced by the spatial resolution of Sentinel-2 (10 m). These findings confirm the potential of YOLO models for aquaculture monitoring through satellite imagery, aligning with recent studies that have implemented variants such as YOLOv8n-HSB or BSSFISH-YOLOv8 to detect structures, fish, or anomalous behaviors in aquatic environments, thus reinforcing their relevance in automated aquaculture analysis.Although accuracy was moderate, the results demonstrate the feasibility of using deep learning with free satellite imagery for aquaculture monitoring, offering a low-cost, scalable, and replicable solution to support environmental management in Lake Titicaca and other high-Andean ecosystems.