Oil spill detection using in Sentinel-1 satellite images based on Deep learning concepts

Awareness of the marine area is very important for crisis management in the event of an accident. Oil spills are one of the main threats to the marine and coastal environments and seriously affect the marine ecosystem and cause political and environmental concerns because it seriously affects the fragile marine and coastal ecosystem. The rate of discharge of pollutants and its related effects on the marine environment are important parameters in assessing the quality of seawater. Effective monitoring, early detection and estimation of the size of these spots are the first and most important step for a successful cleanup operation and that is essential for the relevant authorities to react in a timely manner and limit marine pollution and prevent further damage. Synthetic-aperture radar (SAR) sensors are a very good choice for this purpose due to their effective operation capability regardless of weather conditions and ambient lighting conditions and large area land cover. Black spots related to oil spills can be clearly detected by SAR sensors, but their visual distinction is a challenging goal. The study used artificial aperture radar (SAR) images from the Sentinel-1 satellite to detect oil spills that distributed by European Space Agency (ESA) via the Copernicus Open Access Hub. This paper provides a deep learning framework for identifying oil spills based on a very large data set from around the world, and using the structure of U-Net, DeepLabV3 + and Fc-DenseNet convolutional networks, it classifies images into two classes. In this study, by changing the loss function and deleting single-class images, much better results were obtained than previous similar works. The IoU results for the U-Net, DeepLabV3 +, and FC-DenseNet models were 0.547, 0.513, and 0.545, respectively.