The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Publications Copernicus
Download
Citation
Articles | Volume XLVI-3/W1-2022
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLVI-3/W1-2022, 271–277, 2022
https://doi.org/10.5194/isprs-archives-XLVI-3-W1-2022-271-2022
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLVI-3/W1-2022, 271–277, 2022
https://doi.org/10.5194/isprs-archives-XLVI-3-W1-2022-271-2022
 
22 Apr 2022
22 Apr 2022

NOVEL APPROACHES TO ENHANCE CORAL REEFS MONITORING WITH UNDERWATER IMAGE SEGMENTATION

H. Zhang1, M. Li1,2, X. Pan3, X. Zhang3, J. Zhong1, and J. Qin1 H. Zhang et al.
  • 1State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University, Wuhan, China
  • 2Institute of Theoretical Physics, ETH Zurich, Zurich, Switzerland
  • 3School of Resource and Environmental Sciences, Wuhan University, Wuhan, China

Keywords: Deep Learning, Underwater Image Segmentation, Coral Reefs, Monitoring, Image Processing, Review

Abstract. Coral reefs not only inhabiting millions of species that are primarily or completely associated with them, but also produce economic and cultural benefits to coastal societies around the world. In recent years, affected by climate change and human factors, coral reef ecosystem has been experiencing accelerated degradation. Coral reef monitoring activities are therefore required to assess the impact of adverse factors on corals and to track subsequent recovery or decline. The collection of image data has become a common approach in the field of underwater monitoring, but traditional coral image data analysis mainly has high time and labor costs. We need to investigate the spatial distribution of different coral populations in the study area through image segmentation methods to help oceanographers develop effective management and conservation strategies. In fact, deep learning has shown better prediction performance than traditional image processing or traditional machine learning algorithms in coral image segmentation tasks. Starting from classification of random point annotations, segmentation of sparsely labeled data, and segmentation of densely labeled data, this paper summarizes state-of-the-art techniques of segmentation in deep learning applied to underwater images. Then, we discuss the problems and challenges of CNN-based underwater image segmentation of coral reefs, and make corresponding solutions or possible directions for future.