A 2023 dataset from MLHUB provides Sentinel-2 satellite images focused on marine debris. It supports weakly supervised pixel-level semantic segmentation tasks. The archive includes various sea features and floating materials that co-exist with debris.
Use Cases
- Train weakly supervised semantic segmentation models to classify marine debris pixels in Sentinel-2 imagery.
- Develop models to distinguish floating debris from co-existing features like clear water, turbid water, and waves.
- Classify and segment different floating materials such as Sargassum macroalgae, ships, and natural organic material.
- Benchmark segmentation algorithms on a marine debris-oriented dataset with pixel-level labels.
Strengths
- Focuses on the specific task of marine debris detection from satellite imagery.
- Includes diverse co-existing sea features and floating materials for realistic model training.
Limitations
- Unknown sample size and geographic coverage limit assessment of representativeness.
- Weakly supervised labels may introduce noise compared to fully annotated segmentation datasets.
- Temporal coverage and update frequency are unspecified.
Provenance
- Source
- MLHUB via NASA Earthdata.
- Collection Method
- Curated from Sentinel 2 satellite images.
- Time Range
- null
- Freshness
- Last updated in 2023.
- Geography
- null