GRSG Conference 2020 Presentation

Title: Are Artificial Neural Networks effective for detecting natural oil slicks on SAR imagery?

Author: Cristina Andra Vrinceanu

Synthetic Aperture Radar (SAR) imagery is widely used for the detection and delineation of offshore oil slicks formed through either natural or anthropogenic discharges. This preference is linked to the inherent ability of the radar signal to penetrate cloud, wide-area coverage and a good contrast between dark surfactant slicks and the brighter surrounding sea surface on SAR backscatter images.

Discriminating between areas of high and low backscatter has proven to be a relatively straightforward task for image classification algorithms. In operational scenarios, SAR data is often used in conjunction with ancillary datasets (e.g. wind parameters) to achieve satisfactory results. Throughout time, it has been demonstrated that classification algorithms offer a good degree of accuracy in detecting dark formations.

However, the problem remains in that many false positives and similar structures (also referred as “look-alikes”) are detected, meaning that a final decision for classifying the slick as mineral oil must be made by a trained human operator. This supplementary step can significantly increase the time and resource consumption of the overall process.

With recent hardware advancements, attention has shifted towards the large-scale adoption of deep neural networks for analysing satellite imagery. Convolutional Neural Networks (CNNs) and Generative Adversarial Networks (GANs) have been successfully used for segmenting dark formations and extracting classified petroleum slicks from SAR imagery.

However, these methods require significant computational resources whilst on achieving similar classification accuracies to more the traditional supervised classification algorithms.

In this study, we present the results of the development of an optimized automatic algorithm for natural oil slick detection using open-access Copernicus Sentinel-1 SAR imagery.

This includes a comparison between supervised classification and deep neural network methods. The strengths and shortcomings of each type of technique are analysed in relation to a case study over a set of known seepage sites and potential candidate sites in the Black Sea.