More and more plastic waste is entering the oceans every day. Satellite images can help detect accumulations of litter along coastlines and in the sea and thus eliminate them. A research team has developed a new artificial intelligence model that detects floating plastic in satellite images more accurately than before, even if the images are partially obscured by clouds or hazy weather conditions.

Our society is heavily dependent on plastic products and the amount of plastic waste is expected to increase in the future. If not disposed of or treated properly, much of it accumulates in rivers and lakes. It will eventually flow into the oceans, where it can form deposits of marine debris along with natural materials such as driftwood and algae.

A new study recently published by researchers at Wageningen University and EPFL iSciencedeveloped an AI-based detector that predicts the likelihood of marine debris shown in satellite images. This could help systematically remove plastic waste from the oceans using ships.

Searching satellite images using artificial intelligence

Every 2-5 days around the world, marine debris accumulations can be seen on land and in coastal areas on freely accessible Sentinel-2 satellite images of coastal areas. Since there are terabytes of data, the data needs to be analyzed automatically using artificial intelligence models such as deep neural networks.

“These models were trained on examples provided by oceanographers and remote sensing experts who visually identified thousands of marine debris in satellite images around the world,” says Mark Rooswurm, associate professor at Wageningen University. “So they ‘learned’ a model for plastic waste recognition.”

Improved detection in harsh environments

Researchers have developed an artificial intelligence-based marine debris detector that predicts the probability of marine debris for each pixel in Sentinel-2 satellite images. The detector is trained according to the principles of data-driven AI, which aims to make the most of the limited training data available for the problem.

One example of this is the development of a computer vision algorithm that precisely connects experts’ hand-written descriptions to artifacts visible in images. With this tool, oceanographers and remote sensing experts can provide more samples of training data while being less precise with manually clicking on contours.

Overall, this training method combined with an optimization algorithm trains the AI ​​depth detection model to predict marine debris objects better than previous approaches.

“The detector remains accurate even in more challenging environments, for example when cloud cover and atmospheric haze make it difficult for existing models to accurately identify marine debris,” says Russwurm.

Plastic rubbish following Durban Easter 2019 floods

Detection of plastic in marine litter is particularly important in harsh atmospheric conditions that are cloudy and hazy; because plastic often washes into open water after rain and floods. As the Easter flood in Durban, South Africa, demonstrated, prolonged rains in 2019 caused rivers to overflow and washed away much more garbage than usual.

It was delivered to the open Indian Ocean via the port of Durban. In satellite images, such objects floating among clouds are difficult to distinguish using the usual red-green-blue color “channels”. They can be visualized by switching to other spectral channels, including near-infrared light.

Dual view shows drift directions

In addition to more accurately predicting marine debris clusters, the detection model will also detect debris in daily delivered PlanetScope images.

“Combining weekly Sentinel-2 with daily PlanetScope collections could bridge the gap in continuous daily monitoring,” Rooswurm explained.

“Additionally, PlanetScope and Sentinel-2 sometimes capture the same area of ​​marine debris on the same day, just minutes apart. Dual images of the same object in two locations indicate the direction of drift due to wind and ocean currents in the water. This information is used to improve marine debris drift assessment models.” available.”