This paper introduces a method for offshore oil slick detection. At present, Synthetic Aperture Radar (SAR) is an image acquisition technology useful for oil slick detection in all weather conditions. It is used to carry out the detection, with notable limitations under certain conditions (surfaces, weather conditions). Manual SAR images analysis is expensive and, given the increasing amount of data collected from available sensors, automation becomes mandatory. To achieve this objective, instance object detection relying on deep neural networks is interesting to adapt to the data variability. Relying on such an approach, this article explores the capabilities of generalizing the detection of slicks on large datasets using the Mask-RCNN model. A detailed performance analysis is established in two complementary directions: (i) the impact of the SAR image characteristics(sensor, geographical areas, lookalike presence), (ii) the impact of the neural network architecture, transferred capabilities and training procedures. The main findings of this analysis show that Mask-RCNN features promising performance for pollution detection.

We propose here a generalization of regression trees, referred to as Probabilistic Regression (PR) trees, that adapt to the smoothness of the prediction function relating input and output variables while preserving the interpretability of the prediction and being robust to noise. In PR trees, an observation is associated to all regions of a tree through a probability distribution that reflects how far the observation is to a region. We show that such trees are consistent, meaning that their error tends to 0 when the sample size tends to infinity, a property that has not been established for similar, previous proposals as Soft trees and Smooth Transition Regression trees. We further explain how PR trees can be used in different ensemble methods, namely Random Forests and Gradient Boosted Trees. Lastly, we assess their performance through extensive experiments that illustrate their benefits in terms of performance, interpretability and robustness to noise.