NMFS-OSI/yolo11m-cls-noaa-pacific-benthic-cover-t1

YOLO11m-cls NOAA Pacific Benthic Classifier (Tier-1)

Patch/Point Based Classifer

Model Overview

A YOLO11m image classification model trained to classify broad benthic cover categories (Tier-1) in underwater reef imagery from NOAA Pacific Islands surveys. The model achieves 71.0% top-1 accuracy and 98.8% top-5 accuracy on held-out test data.

• Model Architecture: YOLO11m-cls (medium)
• Task: Image Classification
• Image Size: 224 × 224 pixels
• Classes: 8 broad benthic functional groups

Class	Code	Description
0	CCA	Crustose Coralline Algae
1	CORAL	Hard Coral
2	I	Sessile Invertebrate
3	MA	Macroalgae
4	MF	Mobile Fauna
5	SC	Soft Coral
6	SED	Sediment
7	TURF	Turf Algae

Results & Metrics

Training Performance

Metric	Value
Best Top-1 Accuracy	71.0% (epoch 62)
Best Top-5 Accuracy	98.8% (epoch 63)
Best Validation Loss	0.875 (epoch 62)
Epochs Trained	87 (early stopped)

Test Set Performance

Evaluated on 38,595 ground truth samples:

Metric	Value
Accuracy	70.5%
Balanced Accuracy	64.8%
Top-5 Accuracy	98.8%
Macro F1	0.677
Macro Precision	0.722
Macro Recall	0.648

Per-Class F1 Scores

Class	F1 Score
CCA	0.722
CORAL	0.777
I	0.545
MA	0.624
MF	0.615
SC	0.709
SED	0.832
TURF	0.596

Visualizations

Model Performance Radar

Confusion Matrix

Additional visualizations included:

• confusion_matrix_normalized.png — Training validation confusion matrix
• results.png — Training curves (loss, accuracy, learning rate)

Model Weights

• PyTorch Model: yolo11m_cls_noaa-pacific-benthic-t1.pt
• Best Checkpoint: weights/best.pt

Training Configuration

Parameter	Value
Base Model	yolo11m-cls.pt (pretrained)
Dataset	NMFS-OSI/noaa-pacific-benthic-cover-t1
Training Split	180,087 images
Validation Split	38,589 images
Test Split	38,589 images
Epochs	150 (early stopped at 87)
Patience	25
Batch Size	64
Image Size	224 × 224
Optimizer	AdamW
Initial LR	0.001
LR Schedule	Cosine annealing (cos_lr: true)
Final LR	0.01 × initial
Warmup Epochs	5
Dropout	0.2
Weight Decay	0.01
Precision	AMP (mixed precision)
Seed	42

Augmentations

Augmentation	Value
HSV Hue	0.02
HSV Saturation	0.4
HSV Value	0.3
Rotation	±20°
Translation	0.1
Scale	0.3
Shear	10°
Flip UD	0.5
Flip LR	0.5
Mosaic	1.0

Dataset & Annotations

• Dataset: NOAA Pacific Benthic Cover T1
• Total Images: 257,265 (train/val/test split: 70/15/15)
• Source: NOAA PIFSC Ecosystem Sciences Division (ESD) — NCRMP surveys
• Regions: Hawaii, Marianas, American Samoa, Pacific Remote Island Areas
• Annotation Method: Human analysts using CoralNet interface with Tier-1 functional group labels

How to Use

from ultralytics import YOLO

# Load the trained model
model = YOLO("yolo11m_cls_noaa-pacific-benthic-t1.pt")

# Predict on an image
results = model.predict(source="coral_image.jpg", imgsz=224)

# Get predictions
for result in results:
    top1_idx = result.probs.top1
    top1_conf = result.probs.top1conf.item()
    class_name = result.names[top1_idx]
    print(f"Predicted: {class_name} (Confidence: {top1_conf:.2%})")
    
    # Top-5 predictions
    top5_indices = result.probs.top5
    top5_confs = result.probs.top5conf.tolist()
    for idx, conf in zip(top5_indices, top5_confs):
        print(f"  {result.names[idx]}: {conf:.2%}")

Batch Inference

from ultralytics import YOLO
from pathlib import Path

model = YOLO("yolo11m_cls_noaa-pacific-benthic-t1.pt")

# Predict on folder of images
results = model.predict(source="path/to/images/", imgsz=224)

for result in results:
    print(f"{Path(result.path).name}: {result.names[result.probs.top1]}")

Intended Use

Primary Applications

• Automated benthic cover estimation from underwater survey imagery
• Coral reef monitoring and ecosystem health assessment
• Training baseline for transfer learning to regional datasets
• Research in marine biology and ecosystem science

Out-of-Scope Use

• Species-level identification (model predicts broad functional groups only)
• Regulatory or policy decisions without expert validation
• Regions outside the Pacific without additional fine-tuning
• Low-resolution or poorly-lit imagery may reduce accuracy

Limitations

• Geographic Bias: Trained on Pacific Islands data (Hawaii, Marianas, Samoa/PRIA); may not generalize to Caribbean, Atlantic, or Indo-Pacific regions without fine-tuning
• Class Imbalance: Minority classes (MF, I, SC) have fewer training examples and lower per-class accuracy
• Annotation Uncertainty: Some functional group boundaries are subjective (e.g., turf vs. macroalgae)
• Image Quality: Performance degrades on images with poor lighting, motion blur, or heavy particulates

Ethical Considerations

• Model predictions should supplement, not replace, expert human review for conservation decisions
• Uncertainty estimates should be considered when using predictions for ecological assessments
• Model performance should be validated on local data before deployment in new regions

Environmental Impact

• Hardware: NVIDIA T4 GPU
• Cloud Provider: Google Cloud (us-central1)
• Training Time: ~12 hours

Metadata & Citation

Related Metadata:

• Benthic cover from StRS annotations
• Benthic cover from climate-station annotations
• Benthic images from StRS Sites

Related Resources:

• Dataset: NMFS-OSI/noaa-pacific-benthic-cover-t1
• yolo11m-cls-noaa-pacific-benthic-cover-t3 model

Citation:

Pacific Islands Fisheries Science Center (2025). Ecosystem Sciences Division (ESD); NOAA Fisheries.

Contact

For questions or inquiries:
Michael Akridge — Michael.Akridge@noaa.gov

---

Disclaimer

This repository is a scientific product and is not official communication of the National Oceanic and Atmospheric Administration, or the United States Department of Commerce. All NOAA project content is provided on an 'as is' basis and the user assumes responsibility for its use. Any claims against the Department of Commerce or Department of Commerce bureaus stemming from the use of this project will be governed by all applicable Federal law. Any reference to specific commercial products, processes, or services by service mark, trademark, manufacturer, or otherwise, does not constitute or imply their endorsement, recommendation or favoring by the Department of Commerce. The Department of Commerce seal and logo, or the seal and logo of a DOC bureau, shall not be used in any manner to imply endorsement of any commercial product or activity by DOC or the United States Government.