Deep learning based real time detection of oil residues in freshwater environments using YOLOv5 and YOLOv8

Freshwater ecosystems pose a serious risk due to oil residuals, yet detection and removal remain challenging due to varying illumination, turbidity, and surface disturbances. This work presents a YOLO-based real-time oil-residual detection simulation framework using a set of 1,280 image datasets collected under diverse freshwater conditions. YOLOv5 and YOLOv8 were trained and considered for evaluation using a Roboflow-based preprocessing and augmentation method. The results show that YOLOv5 achieves an mAP@50 (mean Average Precision at 50% IoU) of 88.4%. The YOLOv8 model achieves 91.0% of mAP@50, demonstrating improved precision and localization capability. These findings confirm the potential of the implementation of lightweight deep-learning models for reliable oil-residual detection. It highlights their suitability for YOLO-based models for scalable and adaptable freshwater monitoring and environmental systems.