An efficient coral survey method based on a large-scale 3-D structure model obtained by Speedy Sea Scanner and U-Net segmentation

Abstract

Over the last 3 decades, a large portion of coral cover has been lost around the globe. This significant decline necessitates a rapid assessment of coral reef health to enable more effective management. In this paper, we propose an efficient method for coral cover estimation and demonstrate its viability. A large-scale 3-D structure model, with resolutions in the x, y and z planes of 0.01 m, was successfully generated by means of a towed optical camera array system (Speedy Sea Scanner). The survey efficiency attained was 12,146 m 2 /h. In addition, we propose a segmentation method utilizing U-Net architecture and estimate coral coverage using a large-scale 2-D image. The U-Net-based segmentation method has shown higher accuracy than pixelwise CNN modeling. Moreover, the computational cost of a U-Net-based method is much lower than that of a pixelwise CNN-based one. We believe that an array of these survey tools can contribute to the rapid assessment of coral reefs. Coral reefs play an important role in coastal environments throughout the world, providing food, resources and income to over 500 million people 1 , while supporting up to nine million species and a quarter of all marine life on Earth 2. They also contribute to clean water, removing nitrogen and carbon, and constitute a natural barrier for coastal protection against hurricanes and storms. However, over the last 3 decades, up to 80% of coral cover has been lost in the Caribbean 1 and up to 50% in the Indo-Pacific 3,4 , largely due to anthropogenic stressors that include over-fishing, pollution, sedimentation, habitat destruction and climate change 5-7. An intensive analysis of the extent of coral reef loss and decline in growth was conducted by Pratchett et al. 8. This grave decline requires techniques to rapidly assess coral reef health to enable more effective management and the development of effective conservation strategies 9. Various methods have been developed for monitoring benthic marine habitats such as coral reefs. In general, field transects, such as line intercept transects (LITs), photo line intercept transects (PLITs) and video transects (VTs) have been the most widely used methods, as they are simple to conduct and relatively inexpensive 10-13. However, these in-situ visual methods entail long sampling times due to their small-scale scope, are limited by factors such as diver air tank supply and pose varying degrees of associated risk. To overcome these problems, marine biologists and ecologists have increasingly come to rely on imagery obtained from platforms such as autonomous underwater vehicles (AUVs) or remotely-operated vehicles (ROVs) for marine monitoring 14-17. Such platforms can collect a large number of images, while the total data handling size concurrently increases with technological progress. As a result, much time and effort must be devoted to obtaining ecological data from the collected images, such as the extent of coral reefs and seagrass meadows 18. With recent advancements in computer imaging technologies and growing...

Cost‐effective seafloor habitat mapping using a portable speedy sea scanner and deep‐learning‐based segmentation: A sea trial at Pujada Bay, Philippines

“…During the survey, the P-SSS was set to a depth of 2-6 m from the seafloor, and the boat speed was maintained at 1-2 knots. Here, the altitude affects the visibility and survey coverage (Mizuno et al, 2019(Mizuno et al, , 2020. The fish echosounder installed in the boat monitored the water depth for a safety survey.…”

“…The research team transported the P‐SSS parts in two or three handy suitcases from Japan to the Philippines. In addition, based on the U‐Net architecture (Mizuno et al., 2020; Ronneberger et al., 2015), the research team performed estimations of corals, seagrass and sea urchin covers using large‐scale 2D images (orthophotos) generated from the data obtained by P‐SSS.…”

“…The surveying efficiency of Speedy Sea Scanner is a dramatic improvement over earlier methods (the efficiency of the Speedy Sea Scanner is at least five times greater than that of an AUV and some 80 times higher than that of diving, Mizuno et al, 2020), and a large number of images with greater than before ease can now be obtained.…”

“…In addition, CNN-based semantic segmentation methods for pixel-wise class prediction of input images are widely used for image processing for automatic driving due to their high speed prediction and accuracy (Long et al, 2015;Minaee et al, 2021). As an example of the application of CNNs, in our previous work, the coral distributions for seafloor orthoimages acquired by Speedy Sea Scanner were also successfully estimated (Mizuno et al, 2019(Mizuno et al, , 2020.…”

Mapping of Coral Reefs with Multispectral Satellites: A Review of Recent Papers

“…Coral mapping can be accurately achieved from underwater images, as done in most papers published in 2020 [22][23][24][25][26][27][28][29][30][31][32]. However, a major drawback of underwater images is that they are difficult to acquire at a satisfying time resolution for most remote places, thus making it unfeasible to have a worldwide global map with this kind of data.…”