Publication

Abstract : Oil spills pose a threat and damage marine life and human beings. The illegal discharges of oil spills into the environment create an imbalance in the ecosystem. These oil spills occur due to releasing of oil residues from tankers, ballasts, accidents, or any human interventions. The oil spill detection studies emphasize the classification of oil spill images, detection of dark spots, and separation of trivial objects in oil spill images. The paper presents a study of deep learning technologies such as the traditional state-of-the-art architecture, convolutional neural network (CNN), VGG-16, VGG-19, inception V3, ResNet-50, and Xception to classify the oil spill dataset into two categories–oil spill and non-oil spills using the cross-validation method, Stratified KFold validation. Then, we evaluated the models using the Stratified KFold validation for 5, 10, 15, and 20 splits for 100 epochs, respectively. After performing the experiments, the ResNet-50 model gave the best score of 100% consistently for 5, 10, 15, and 20 splits, respectively.