Real-Time Oil Spill Detection with YOLO Framework for Marine Ecosystem Surveillance

Abstract

Early detection of oil spills is crucial and essential for marine environments to minimize environmental harm and enable quick responsive measures. Oil spills can cause significant ecological and financial losses, which emphasizes the need for an efficient monitoring system. This paper presents the use of YOLO deep learning algorithms to enhance the oil spill detection speed and accuracy. A robust and high-quality dataset is taken, consisting of images extracted from Roboflow. To maximize the data quality, preprocessing techniques such as label normalization, contrast enhancement and noise reduction were used. The proposed YOLO algorithms were trained using Adam and SGDM optimizers with an initial learning rate of 0.01, 0.001 and 0.0001. Among the adopted YOLO models, YOLOv9 yielded impressive results with an mAP@0.5 of 94.45%, precision of 95.6%, recall of 93.3% and F1 Score of 94.44%. The recommended system, which incorporates deep learning technologies into marine environment monitoring, greatly improves the marine surveillance systems for oil spill detection and emergency response capabilities by enabling real-time monitoring.