MARIDA Dataset Card - elena-andreini/TriesteItalyChapter_PlasticDebrisDetection GitHub Wiki

MARIDA Dataset Card

Dataset Name:

MARIDA (MARIne Debris Archive)

Description:

Marine Debris Archive (MARIDA) is a marine debris-oriented dataset on Sentinel-2 satellite images. It also includes various sea features (clear & turbid water, waves, etc.) and floating materials (Sargassum macroalgae, ships, natural organic material, etc) that co-exist. MARIDA is primarily focused on the weakly supervised pixel-level semantic segmentation task. The dataset provides pixel-wise segmentation for various marine objects and covers a broad temporal range, making it one of the most comprehensive datasets for floating plastic debris detection.

Key Characteristics:

Satellite Mission: Sentinel-2
Data Type: The MARIDA dataset primarily uses GeoTiff format for storing the satellite imagery patches and their corresponding masks

Imagery Level: Level-1C

Atmospheric Correction:

  • Applied: Yes
  • Algorithm: ACOLITE

Bands : 11 Spectral Bands (B1, B2-B7, B8, B8A, B11, B12) are used. B9 and B10 are not used

Spatial Resolution: 60m (B1), 10m (B2, B3, B4, B8), 20m (B5, B6, B7, B8A, B11, B12)

Temporal Coverage: 2015-2021

Geographic Coverage: Global including :

  • C.America/Guatemala (Sentinel-2 tile 16PCC)
  • C.America Honduras (Sentinel-2 tiles 16PEC, 16QED)
  • N.America/S.Domingo (Sentinel-2 tiles 19QDA)
  • N.America/Haiti (Sentinel-2 tiles 18QWF/QYF, QYG)
  • Asia/Indonesia (Sentinel-2 tiles 50LLR)
  • Asia/Vietnam (Sentinel-2 tile 48PZC)
  • Asia/Philippines (Sentinel-2 tile 51PTS)
  • Europe/Scotland (Sentinel-2 tile 30VWH)
  • Africa/South Africa (Sentinel-2 tile 36JUN)
  • Asia/South Korea (Sentinel-2 tile 52SDD)
  • Asia/Indonesia (Sentinel-2 tile 48MXU/MYU)
  • Asia/China (Sentinel-2 tile 51RVQ)

Labels/Annotations:

  • Type: Pixel-wise segmentation
  • Classes:
    1. Marine Debris: This is the primary focus of the dataset, encompassing various types of floating debris.
    2. Sargassum macroalgae: A type of seaweed that can be mistaken for debris.
    3. Ships: Distinguishing vessels from other features.
    4. Natural Organic Material: Other organic matter that can float on the surface.
    5. Waves: Sea surface disturbances.
    6. Wakes: Disturbances left by ships or other vessels.
    7. Foam: Created by waves and other surface activity.
    8. Dissimilar water types: Includes clear, turbid, sediment-laden, and shallow water.
    9. Clouds: Distinguishing cloud cover from other features.

(1: Marine Debris, 2: Dense Sargassum, 3: Sparse Sargassum, 4: Natural Organic Material, 5: Ship, 6: Clouds, 7: Marine Water, 8: Sediment-Laden Water, 9: Foam, 10: Turbid Water, 11: Shallow Water, 12: Waves, 13: Cloud Shadows, 14: Wakes, 15: Mixed Water)

Debris-Free Scenes: Yes

Relevance to Project:

  1. It provides high-resolution Sentinel-2 imagery with detailed pixel-wise annotations, allowing for accurate plastic debris detection.
  2. It includes both debris-free and polluted scenes, enabling robust training of machine learning models to distinguish between plastic debris and other floating materials.
  3. MARIDA integrates varied marine conditions (turbidity levels, seasonal variations, atmospheric changes), making it ideal for developing generalizable models.
  4. Supports validation of index-based methods (e.g., NDPI, FDI, floating debris indices) and deep learning models.
  5. Facilitates comparisons between remote sensing techniques and in-situ observations, which can enhance detection accuracy in satellite-based monitoring.

Access: https://source.coop/ntua/marida

Additional Notes:

  1. Data is processed with atmospheric corrections to reduce noise and enhance classification accuracy.
  2. Includes varied environmental conditions such as high turbidity, wave action, and different lighting conditions.
  3. Works well with multi-modal datasets, including UAV imagery, for high-resolution validation.
  4. One of the largest marine debris datasets for Sentinel-2 imagery, with extensive annotations.

Next Steps:

  1. Filter data specifically for plastic debris classification.
  2. Apply preprocessing techniques such as cloud masking, spectral filtering, and re-projection.