The Suitability of Machine-Learning Algorithms for the Automatic Acoustic Seafloor Classification of Hard Substrate Habitats in the German Bight
-
Published:2023-08-21
Issue:16
Volume:15
Page:4113
-
ISSN:2072-4292
-
Container-title:Remote Sensing
-
language:en
-
Short-container-title:Remote Sensing
Author:
Breyer Gavin1ORCID, Bartholomä Alexander1, Pesch Roland2
Affiliation:
1. Senckenberg am Meer, Marine Research, Südstrand 40, 26382 Wilhelmshaven, Germany 2. Jade Hochschule Oldenburg, Institute for Applied Photogrammetry and Geoinformatics (IAPG), Ofener Str. 16/19, 26121 Oldenburg, Germany
Abstract
The automatic calculation of sediment maps from hydroacoustic data is of great importance for habitat and sediment mapping as well as monitoring tasks. For this reason, numerous papers have been published that are based on a variety of algorithms and different kinds of input data. However, the current literature lacks comparative studies that investigate the performance of different approaches in depth. Therefore, this study aims to provide recommendations for suitable approaches for the automatic classification of side-scan sonar data that can be applied by agencies and researchers. With random forests, support vector machines, and convolutional neural networks, both traditional machine-learning methods and novel deep learning techniques have been implemented to evaluate their performance regarding the classification of backscatter data from two study sites located in the Sylt Outer Reef in the German Bight. Simple statistical values, textural features, and Weyl coefficients were calculated for different patch sizes as well as levels of quantization and then utilized in the machine-learning algorithms. It is found that large image patches of 32 px size and the combined use of different feature groups lead to the best classification performances. Further, the neural network and support vector machines generated visually more appealing sediment maps than random forests, despite scoring lower overall accuracy. Based on these findings, we recommend classifying side-scan sonar data with image patches of 32 px size and 6-bit quantization either directly in neural networks or with the combined use of multiple feature groups in support vector machines.
Funder
Federal Ministry of Education and Research
Subject
General Earth and Planetary Sciences
Reference74 articles.
1. Galvez, D., Papenmeier, S., Sander, L., Hass, H., Fofonova, V., Bartholomä, A., and Wiltshire, K. (2021). Ensemble Mapping and Change Analysis of the Seafloor Sediment Distribution in the Sylt Outer Reef, German North Sea from 2016 to 2018. Water, 13. 2. Zhao, T., Montereale Gavazzi, G., Lazendić, S., Zhao, Y., and Pižurica, A. (2021). Acoustic Seafloor Classification Using the Weyl Transform of Multibeam Echosounder Backscatter Mosaic. Remote Sens., 13. 3. Mapping seabed sediments: Comparison of manual, geostatistical, object-based image analysis and machine learning approaches;Diesing;Cont. Shelf Res.,2014 4. Combining pixel and object based image analysis of ultra-high resolution multibeam bathymetry and backscatter for habitat mapping in shallow marine waters;Ierodiaconou;Mar. Geophys. Res.,2018 5. Misiuk, B., Diesing, M., Aitken, A., Brown, C.J., Edinger, E.N., and Bell, T. (2019). A Spatially Explicit Comparison of Quantitative and Categorical Modelling Approaches for Mapping Seabed Sediments Using Random Forest. Geosciences, 9.
|
|