Oil Identification on Beaches Using Deep Learning Techniques

The oil spill on the beaches of the Brazilian northeast coast has caused disastrous consequences, both for us human beings, as well as for the marine biodiversity. Bearing in mind that its identification is not easy, we propose to do it in this article, using a machine learning approach, a type of artificial intelligence widely used in the area of image classification, alongside with a Convolutional Neural Network. In this regard, we have as main objective the training of a model able to distinguish images, automatically, in three types of classes: (i) normal beaches; (ii) beaches with gulf-weed, a biological indicator of pollution; and (iii) beaches with oil. In the best scenario evaluated, we achieved an average accuracy of 91%.