MARIDA: A benchmark for Marine Debris detection from Sentinel-2 remote sensing data.

Area of Science:

• Remote Sensing and Earth Observation
• Marine Science and Oceanography
• Artificial Intelligence and Machine Learning

Background:

• Growing research focuses on remote sensing for marine debris detection and spectral analysis.
• Existing methods require robust datasets for developing operational monitoring solutions.
• Distinguishing marine debris from similar oceanic features is a significant challenge.

Purpose of the Study:

• To introduce the Marine Debris Archive (MARIDA), a benchmark dataset for developing and evaluating Machine Learning (ML) algorithms for marine debris detection.
• To provide a comprehensive dataset based on Sentinel-2 multispectral satellite data.
• To enable the research community to develop and evaluate AI-driven solutions for marine debris monitoring.

Main Methods:

• Creation of MARIDA, the first dataset utilizing Sentinel-2 satellite data for marine debris detection.
• Inclusion of georeferenced annotations (polygons/pixels) for verified plastic debris events across diverse global conditions (seasons, years, sea states).
• Presentation of spectral and statistical analysis of the dataset, alongside ML baselines for semantic segmentation and multi-label classification.

Main Results:

• MARIDA dataset successfully distinguishes marine debris from co-existing marine features like macroalgae, ships, foam, and various water types.
• Established ML baselines demonstrate the dataset's utility for weakly supervised semantic segmentation and multi-label classification tasks.
• The dataset covers diverse geographical regions and environmental conditions, enhancing the generalizability of developed models.

Conclusions:

• MARIDA is an open-access resource crucial for advancing AI and deep learning-based marine debris detection.
• The dataset facilitates exploration of spectral characteristics of floating materials and sea state features.
• MARIDA supports the development of improved satellite pre-processing pipelines and operational monitoring systems for marine pollution.