Publication

Abstract : Marine ecosystems are endangered by oil spills, so quick detection is vital for effective response. This paper introduces an automated, portable oil spill detection system that uses the You Only Look Once Version 8 (YOLOv8) segmentation model (YOLOv8-seg). First, the study evaluates and compares YOLOv5, YOLOv8-seg, U-Net, and Mask RCNN on the dataset from RobOflow. We look at the trade-offs between segmentation accuracy and computational efficiency. Mask R-CNN and U-Net provide higher boundary precision but have slow inference speeds, which restrict their use in real-time situations. In contrast, YOLOv8-seg is 79% faster than U-Net and 91.6% faster than Mask R-CNN, achieves a favorable balance between accuracy and inference speed, making it suitable for edge computing applications. The model is deployed on the low power Grove Vision AI module, which demonstrates real-time, on-site oil spill detection capabilities. This setup enables immediate data processing, enabling quicker environmental responses and supporting ecological monitoring efforts. The findings suggest that YOLOv8-seg offers an effective balance between precision and efficiency, making it a practical solution for automated marine oil spill monitoring in resource-limited, field-ready contexts.