Exploration and Evaluation of Machine Learning Methods for Detecting Oil Spills in Open Seas

Detecting oil spills in open seas is important for the early containment of the pollution diffusion and avoidance of further ecological catastrophes. Herein, we explore and assess the application of several machine learning (ML) methods and image processing techniques for the detection of oil spills in images of open seas captured at the elevation of planes, helicopters, or drones. The image dataset used contains 24 clean sea images and 24 dirty sea images with oil slicks. The machine learning methods experimented with include neural networks, SVM, random forest, decision tree, naïve Bayes, AdaBoost, and logistic regression. We also evaluated these ML methods with and without the application of Sobel Feldman edge detection and Vintage filtering image processing techniques. Results indicate the superiority of the in and neural networks methods over the other explored ML methods, achieving perfect and near-perfect classification, respectively.