Leveraging Deep Learning Techniques for Marine and Coastal Wildlife Using Instance Segmentation: A Study on Galápagos Sea Lions

Instance segmentation is a powerful deep learning technique that goes beyond traditional object detection by identifying and delineating individual objects within an image. This study evaluates several algorithms for instance segmentation of sea lions from video frames captured in the Galápagos Islands. The YOLO v8 and v9 models demonstrated superior metric values compared to other models, including those based on Detectron2 with a Mask R-CNN backbone. The study addressed challenges such as species proximity during group activities, variability in lighting conditions and image quality. YOLO v8 achieved an average precision (AP50) rate of 97.90%, while YOLO v9 achieved a precision rate of 98.2% for sea lion instance segmentation. These results provide valuable resources for future analysis of wildlife monitoring methods across diverse natural environments, with significant implications for conservation.