Oil Spill Identification using Deep Convolutional Neural Networks

Oil spill detection is an extremely important topic in which Machine Learning (ML) can be utilized because oil spills that go undetected can cause huge environmental negative impacts. The science of how an oil spill can cause devastation to wildlife has been widely viewed with sorrow and the early detection of oil spills can greatly reduce the negative impact oil spills have on the environment. If an optimal solution is found for oil spill detection, continuous monitoring can be achieved either through satellite images or images obtained from unmanned aerial vehicles such as drones. In this paper, we develop a dataset for oil spill detection collected from images from the Internet and other online resources. The dataset consists of 783 images of Oil Spills and 783 normal images. Since these are real-world images, research done on this dataset will produce a more realistic and practical solution. In this paper, we also propose an enhanced CNN model based on GoogleNet and VGG16 combined with transfer learning for the detection and classification of oil spills. The GoogleNet Transfer Learning model achieved better results of training accuracy of 97.5%, training loss of 0.0894, and validation accuracy of 95.6%. Since this is a new dataset, the results cannot be compared to anything in the extant literature.