Oil spills pose severe threats to marine and coastal ecosystems, requiring fast and accurate detection methods for timely response. Existing approaches often struggle with generalizing across heterogeneous environments or achieving high precision in real-time applications. This work investigates the application of YOLOv9, a state-of-the-art one-stage object detector, to oil spill detection in maritime and terrestrial contexts. High-resolution drone images were prepared into Water Only, Land Only, and Combined Water-Land datasets with extensive preprocessing and augmentation. YOLOv9 models were trained and evaluated by measuring precision, recall, and mAP. The Combined Water-Land dataset achieved the highest precision at 74.5%, outperforming the Water-Only (72.6%) and Land-Only (65.6%) datasets, demonstrating superior generalization across diverse environmental scenarios. Calibration curves indicated excellent convergence and stable training, with mAP@50 values consistently above 0.75, confirming robust detection performance. This research demonstrates that YOLOv9 dramatically improved the speed of detecting oil spills in real-time heterogeneous scenarios as compared to other methods. This method enables operational oil spill monitoring from aerial imagery in an effortless and dependable manner.