CSIRO Patch-based 8-class Morphology Model
A patch-based segmentation model designed for seagrass morphological class identification and habitat mapping. The model segments images into fixed-size patches (e.g., 280×280 pixels) and classifies each patch into foreground classes (various seagrass species) or background classes (such as substrate, coral, and macroalgae). Morphological seagrass coverage is quantified by calculating the percentage of patches classified into each class, enabling precise assessments of seagrass distribution and density.
Input
The model is trained on images from Indo-Pacific regions, including Indonesia, Thailand, and Fiji. While it can process images from other regions, accuracy may decrease.
Classes
This model classifies seagrass into the following morphological categories:
- Oval Seagrass - Ho, Hd, Hb
- Cylindrical Seagrass - Si
- Stemmy Seagrass - Tc
- Strappy Medium/Thin Seagrass - Cs, Th, Hu, Cr, Zc
- Strappy Hair-Thin Seagrass - Hp
- Strappy Thick Seagrass - Ea
- Unknown or Mixed Seagrass
- Others
Note: The morphological codes in italics 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 50x50 cm quadrat captured by a camera positioned at a fixed distance. The images are resized to 3008x3008 pixels, achieving a resolution of approximately 60 pixels per cm.
Patch Size
280x280 pixels: Corresponds to a 5x5 cm area.
Annotated Patches on Reefcloud
Due to software limitations, annotators are required to label 20 patches per image slice, arranged in a 5x4 grid. This setup allows for a maximum of 1024 patches (32x32 grid) per image.
Input Standardisation
Every image is resized and split into 100 (10x10) 280x280px patches as input to the ML model.
Patch Labeling Methodology
Single Seagrass Species Present: Assign the specific seagrass class label to the patch, regardless of the percentage of coverage within the patch.
Multiple Seagrass Species Present:
- If one species is clearly dominant, label the patch with the dominant species’ class.
- If no species is dominant, use a generic seagrass class label.
- Seagrass vs. Benthos Classes: Seagrass classifications take precedence over benthic classes. If no seagrass is visible, label the patch with the dominant benthic class.
Contextual Considerations
Annotators focus on the content within the patch while also considering the broader context of the entire image. This approach ensures that local features are interpreted in alignment with the global scene, leading to more accurate and meaningful annotations.
Results Interpretation
For each input patch from an image, the ML model will predict the majority seagrass class within the patch if any seagrass is present, otherwise it will be classified as one of the non-seagrass classes.
The image-level percentage cover estimate is an aggregated percentage of image patches results. For example, if there are 20 patches out of 100 classified as Stemmy Seagrass, the image-level results will show 20% for Stemmy Seagrass. Note that the image-level estimates are an approximation to the human-based visual survey and are generally leaning towards over-estimation for dominant classes.
Model Metadata
Version: 1.0
Tags:
Credits: Yang Li, Rizwan Khokher, Brendan Do, Asheley Stacey, Jeremy Oorloff, Jiajun Liu, Brano Kusy
Performance on Test Dataset
| 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
