Annotation Tool
To assist in quantifying benthic biodiversity in marine restoration projects, CLIMAREST has developed an annotation tool, which uses AI and deep learning to classify, detect, and segment benthic ecosystems from data across relevant environments. The tool utilizes all available information across different underwater remote sensing modalities, aggregating different observational, satellite, LiDAR, underwater remote sensing, and in-site datasets. This annotation tool goes beyond contemporary monitoring techniques by merging all these datapoints and sets with automated data classification, which increases data processing efficiency.
Svalbard Annotation Tool Demonstration
This demonstration provides an overview of of how to implement AI image classification algorithms, utilize the CLIMAREST annotated dataset, and how to register and use Squidle+. The demo also illustrates an analysis for the georeferenced imagery dataset collected by the Blueye X3 in June 2024 from the Adventfjorden as part of the efforts to study the impact of untreated sewage in the fjord. It also has a demo of how AI models could be used to classify images with visual debris as well as estimating the area of objects of interest in the images.
Background on the Use of ROV Technology in the Svalbard Demo Site
Arctic marine ecosystems are highly vulnerable to external stressors including to global climate change and local anthropogenic disturbances such as pollution. The municipality of Longyearbyen, Norway (located in the high Arctic) installed a sieve in May 2022 for mechanical treatment of wastewater to mitigate discharge of macro-pollutants to the fjord. The nearby research community Ny-Ålesund had a biological treatment facility installed in 2018. Previously, raw sewage was directly pumped into the fjord, disturbing the natural ecosystem. As part of the CLIMAREST project, researchers used a Remotely Operated Vehicle (ROV) to collect images of the seafloor to detect anthropogenic debris near the sewage outlet. These images were used to monitor the effectiveness of the installed sieve and a social campaign run in Longyearbyen that aimed to educate locals and tourists about correct disposal of waste.
In this demonstrator, we can have a look at the georeferenced images collected by the Blueye X3 ROV at different locations in the Adventfjorden. We can also look at how AI models can help us in the automatic classification of images with visual anthropogenic debris as well as estimating area of debris/objects of interest in the images. The annotated image dataset is openly accessible through squidle.org
