Marine Debris Object Detection Dataset from PlanetScope Satellite Imagery

256x256 pixel satellite image chips with bounding box labels for floating marine debris objects. The dataset, created by MLHUB and hosted on NASA EarthData, was last updated in 2023. It focuses on debris types including plastics, algae, sargassum, wood, and other artificial items.

Use Cases

• Train object detection models to identify bounding boxes for plastics, algae, and other debris classes in 256x256 pixel image chips.
• Develop geospatial analysis tools using the provided geographical coordinates to map debris locations and concentrations.
• Create multi-class classifiers to distinguish between natural materials like sargassum and artificial items like plastics within the bounding boxes.
• Benchmark model performance on satellite imagery with 3-meter spatial resolution for environmental monitoring applications.

Strengths

• Images sourced from PlanetScope optical imagery with approximately 3-meter spatial resolution.
• Labels include bounding boxes and geographical coordinates for precise localization.
• Debris is classified into specific categories: plastics, algae, sargassum, wood, and other artificial items.

Limitations

• Geographic coverage is biased, with most observations from the Bay Islands in Honduras and only a small percentage from Ghana and Greece.
• The dataset size, row count, and class distribution balances are unknown, which may affect model training.
• Data recency is unclear beyond the last platform update in 2023.

Provenance

Source

NASA EarthData, contributed by MLHUB.

Collection Method

Images obtained from PlanetScope optical satellite imagery, with labels from several studies for collection and validation.

Time Range

null

Freshness

Last updated on the platform in 2023-01-01.

Geography

Primarily the Bay Islands in Honduras, with minor coverage of coastlines in Ghana and Greece.