Use of semantic segmentation for mapping Sargassum on beaches
Javier Arellano-Verdejo1, Martin Santos-Romero2, Hugo E Lazcano-Hernandez3
1Department of Observation and Study of the Earth, Atmosphere and Ocean, El Colegio de la Frontera Sur, Chetumal, Quintana Roo, Mexico.
Peerj
|June 15, 2022
Summary
This study introduces a new method using semantic segmentation of geotagged photos to accurately map Sargassum seaweed coverage on beaches. This approach achieves 91% accuracy, improving previous estimations.
Area of Science:
- Marine Biology
- Remote Sensing
- Geographic Information Systems (GIS)
Background:
- Unusual influxes of Sargassum seaweed present significant challenges for Caribbean beaches.
- Accurate monitoring and estimation of Sargassum coverage are complex and ongoing issues.
Purpose of the Study:
- To develop and validate a novel mapping methodology for estimating Sargassum coverage on beaches.
- To improve the accuracy of Sargassum beach coverage quantification.
Main Methods:
- Utilized semantic segmentation of geotagged photographs to create detailed maps.
- Developed the first dataset of segmented Sargassum images for model training.
- Employed a crowdsourcing scheme for data collection.
Main Results:
- The proposed mapping methodology achieved an accuracy of 91%.
- This represents a significant improvement over state-of-the-art methods.
- The method accurately visualizes percent coverage, unlike presence/absence data.
Conclusions:
- Semantic segmentation of geotagged images offers a highly accurate solution for mapping Sargassum coverage.
- This methodology enhances the ability to monitor and manage Sargassum events.
- The developed dataset and model contribute to addressing the Sargassum problem.
