Researchers at Nanjing University have created a global dataset of satellite images to improve how image recognition models detect oil pollution on the ocean’s surface. The dataset includes over 200,000 images captured by the Sentinel-1 satellite between 2014 and 2020—half showing confirmed oil slicks, and half showing visually similar features like algae, calm water, and other natural phenomena that often get mistaken for oil in satellite imagery. Training image recognition models on both types of data significantly boosted their real-world performance: detection accuracy rose from 56 percent to as high as 96 percent.
Image Credits: Brian Kyed