Using Convolutional Neural Network for the Detection of Offshore and Onshore Oil Spills

Abstract Oil spills pose significant environmental, economic, and health challenges, necessitating effective monitoring and detection methods. Traditional techniques, such as field investigations and remote sensing, are costly and often lack the speed and scalability required for real-time detection. This study proposes a novel approach for oil spill detection using convolutional neural networks (CNNs) deployed on edge devices, specifically mobile phones. Leveraging pre-trained models such as EfficientNet-Lite1, MobileNetV2, and ResNet50, the prediction layers were fine-tuned using domain-specific data. The EfficientNet-Lite1 model achieved the highest accuracy of 99.9%, followed closely by MobileNetV2 with 99.1%. Post-training quantization techniques, such as float16 quantization, were applied to optimize model size and inference speed. The best-performing model was exported as a TensorFlow Lite model and integrated into an Android mobile application developed using Flutter. The app, powered by the ML-Kit library, demonstrated real-time oil spill detection with high accuracy and efficiency. This study highlights the potential of edge devices to run sophisticated deep learning models, eliminating the need for costly IT infrastructures. The proposed solution offers a scalable, cost-effective, and user-friendly approach to oil spill detection, with the potential to revolutionize environmental monitoring and response systems. Future research will focus on expanding data diversity, geographical coverage, and integrating the technology into other edge devices like drones and smartwatches for enhanced applicability.