Looking for Seagrass: Deep Learning for Visual Coverage Estimation

Underwater videography enables marine researchers to collect enormous amounts of seagrass image data. This collection is fast and cheap but the manual analysis of such data is slow and expensive. Therefore, we propose a machine-learning approach for the automatic seagrass coverage estimation of the...

Full description

Saved in:
Bibliographic Details
Published in2018 OCEANS - MTS/IEEE Kobe Techno-Oceans (OTO) pp. 1 - 6
Main Authors Reus, Gereon, Moller, Thomas, Jager, Jonas, Schultz, Stewart T., Kruschel, Claudia, Hasenauer, Julian, Wolff, Viviane, Fricke-Neuderth, Klaus
Format Conference Proceeding
LanguageEnglish
Published IEEE 01.05.2018
Subjects
Clustering algorithms
Convolutional neural networks
Feature extraction
Image segmentation
Logistics
Measurement
Training
Online AccessGet full text

Cover

More Information
Summary:Underwater videography enables marine researchers to collect enormous amounts of seagrass image data. This collection is fast and cheap but the manual analysis of such data is slow and expensive. Therefore, we propose a machine-learning approach for the automatic seagrass coverage estimation of the sea bottom. Our contribution is the investigation of CNN features to describe patches and superpixels of seagrass. CNN features are the activations of a specific layer in a deep convolutional neural network. We also provide the first public available dataset of seagrass images that can be used as a benchmark for automatic seagrass segmentation. Our best method achieves an accuracy of 94.5% for seagrass segmentation on the provided dataset. Our code and dataset is available on GitHub: https://enviewfulda.github.io/LookingForSeagrass/.
DOI:10.1109/OCEANSKOBE.2018.8559302