Model Garden

CSIRO Patch-based 8-class Morphology Model (Indo-Pacific)

This model is a patch-based coarse image segmentation framework designed for seagrass morphological identification and marine habitat mapping across eight distinct classes: seven seagrass morphology classes and one background class. Compared with the Australian model, it excludes the ferny morphology class because this type, specifically Halophila spinulosa, is rare or absent in the target Indo-Pacific regions. The framework processes high-resolution benthic imagery by dividing each image into fixed-size patches (e.g., 280×280 pixels), and classifying each patch into a foreground seagrass morphology class, such as strappy or stemmy, or into a single background category, labelled as others. By aggregating patch-level predictions, the model estimates morphology-level seagrass coverage as a percentage of the total image area, providing a scalable and automated tool for seagrass distribution and density assessment.

Try this model
Example image classified by CSIRO Patch-based 8-class Morphology Model (Indo-Pacific)

Details

Input

The model is trained on benthic seagrass images collected from Indo-Pacific regions, including Indonesia, Thailand, and Fiji. Although the model can process images from other regions, prediction accuracy may decrease when applied to imagery with different environmental conditions, camera settings, or seagrass communities.

Classes

This model classifies image patches into the following seagrass morphological categories:

  • Oval Seagrass — Ho, Hd, Hb: Halophila ovalis, Halophila decipiens, Halophila beccarii
  • Cylindrical Seagrass — Si: Syringodium isoetifolium
  • Stemmy Seagrass — Tc: Thalassodendron ciliatum
  • Strappy Medium/Thin Seagrass — Cs, Th, Hu, Cr, Zc: Cymodocea serrulata, Thalassia hemprichii, Halodule uninervis, Cymodocea rotundata, Zostera capricorni
  • Strappy Hair-Thin Seagrass — Hp: Halodule pinifolia
  • Strappy Thick Seagrass — Ea: Enhalus acoroides
  • Unknown or Mixed Seagrass
  • Others

Note: Species codes correspond to specific seagrass species. For full species names and descriptions, refer to the Seagrass-Watch Species ID Guide.

Annotation Method

Each image represents a 50 × 50 cm quadrat captured by a camera positioned at a fixed distance from the substrate. Images are resized to 3008 × 3008 pixels, giving an approximate resolution of 60 pixels per cm.

Patch Size

Each patch is 280 × 280 pixels, corresponding to an approximate 5 × 5 cm area.

Annotated Patches on ReefCloud

Due to software limitations, annotators label 20 patches per image slice, arranged in a 5 × 4 grid. This setup allows up to 1024 patches per image, arranged as a 32 × 32 patch grid.

Patch Labelling Methodology

If a patch contains a single seagrass species or morphology, assign the corresponding seagrass morphology label, regardless of the percentage cover within the patch.

If multiple seagrass species or morphologies are present, assign the label of the clearly dominant seagrass type. If no type is clearly dominant, use the Unknown or Mixed Seagrass label.

Seagrass labels take precedence over non-seagrass benthic classes. If no seagrass is visible, label the patch as Others, based on the dominant non-seagrass content.

Contextual Considerations

Annotators should focus primarily on the content within each patch while also considering the broader context of the full image. This helps ensure that local patch-level features are interpreted consistently with the surrounding scene, leading to more accurate and meaningful annotations.

Performance on Test Dataset

Accuracy

Class Group Accuracy Support
Others 0.7626 2,759
S_CY(Si) 0.0714 14
S_OV(Ho Hd Hb) 0.5000 6
S_STM(Tc) 0.7582 335
S_STP_M&N(Cs Th Cr) 0.5079 5,475
S_STP_N(Hp) 0.5294 272
S_STP_T(Ea) 0.5558 2,134
S_UKN 0.0635 945

Confusion Matrix

Example Output

Example Input