Context:
This project developed an automated coral detection and classification model using YOLO. By processing video clips, the model can recognize and classify different coral species in real time, providing a scalable and efficient alternative to manual annotation. This approach improves the accuracy and speed of coral data collection and can aid long-term research and large-scale monitoring efforts.
Data:
Imagery from this project came from the National Oceanic and Atmospheric Administration (NOAA) Office of Ocean Exploration and Research (OER). The imagery was capture on DATE by the Insite Pacific Zeus Plus Camera on the ROV Deep Discoverer from Keahole coral bed off the Kona Coast of the Big Island of Hawaii at 380 m. The video used in this project was archived and made available through Ocean Networks Canada's Oceans 3.0 digital infrastructure and can be accessed at https://doi.org/10.34943/a25f1b96-a344-438f-989f-7c38701634cb Ocean Networks Canada Society. 2016. Deep Discoverer Remotely Operated Vehicle Camera Deployed 2015-01-01. Ocean Networks Canada Society. https://doi.org/10.34943/a25f1b96-a344-438f-989f-7c38701634cb. Data set contains about 800 annotated images in size 1024x1024. Method:
For dataset preparation, the video was split into frames, and the training dataset was annotated with bounding boxes and species labels for training and evaluation in Roboflow, focusing on three coral species: Gorgonian Coral, Bamboo Coral, and Black Coral. YOLOv11 was selected for object detection and classification, as it is the most stable among recent releases. During video processing, each frame was analyzed using the YOLO model, and ByteTrack was applied to maintain consistent IDs for coral instances across frames, ensuring accurate counting. In post-processing, a summary report was generated, visualizing coral species distribution and counts using a stacked bar chart.
Evaluation:
Precision-Recall Curve:
Black Coral has the best PR curve with an AP of 0.902, indicating a high quality classification for this category. Bamboo Coral performed moderately well with an AP of 0.723. Gorgonian Coral was the least well recognized, with an AP of 0.642, indicating that the model's predictions for this category were not stable. The mean value for all categories mAP at 0.5 IoU is 0.756, indicating that overall the model performed reasonably well, but could be further optimized.
Normalized Confusion Matrix:
Diagonal elements indicate the percentage of correct classifications. Black Coral had the highest accuracy, 91% of black coral samples were correctly classified. Bamboo Coral has an accuracy of 78%. Gorgonian Coral was relatively poorly classified (76%) and had a high number of misclassifications as background (69%).
Model use-case:
Hypothesis: Coral populations in deep-sea habitats exhibit seasonal variations in species distribution and abundance due to environmental factors like temperature, ocean currents, and nutrient availability. My model can accurately identifies coral species, reducing manual annotation workload. Processes thousands of video frames quickly, making long-term monitoring feasible. Also can be expanded to monitor other marine species and ecosystem changes.
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.