The ocean is changing at an unprecedented rate, making it difficult to maintain accountable stewardship while visually monitoring vast amounts of marine data. The amount and pace of data collection required is outpacing our ability to rapidly process and analyze it as the research community searches for baselines. Lack of data consistency, inappropriate formatting, and the desire for meaningful, labeled data sets have contributed to the limited success of recent advances in machine learning, which have enabled faster and more complex visual analysis of data.
To meet this requirement, several research institutions worked with MBARI to accelerate ocean research using the capabilities of artificial intelligence and machine learning. One of the results of this partnership is FathomNet, an open source image database that employs state-of-the-art data processing algorithms to standardize and aggregate carefully selected labeled data. The team believes that the use of artificial intelligence and machine learning will be the only way to speed up critical studies on the health of the oceans and remove the bottleneck for underwater image processing. Details about the development process behind this new image database can be found in a recent research publication in the journal Scientific Reports.
Machine learning has historically transformed the field of automated visual analysis, thanks in part to the large volumes of annotated data. When it comes to terrestrial applications, the go-to data sets that machine learning and machine vision researchers turn to are ImageNet and Microsoft COCO. To provide researchers with a rich and engaging standard for underwater visual analysis, the team created FathomNet. In order to establish a highly maintained and freely accessible underwater imagery training resource, FathomNet combines images and recordings from many different sources.
Researchers at MBARI’s Video Lab carefully annotated data representing nearly 28,000 hours of deep-sea video and more than 1 million deep-sea photographs that MBARI collected over 35 years. Around 8.2 million annotations documenting observations of animals, ecosystems and objects are present in the MBARI video library. This comprehensive data set serves as an invaluable tool for institute researchers and their international collaborations. The National Geographic Society’s Exploration Technology Laboratory collected more than 1,000 hours of video data from various marine habitats and locations throughout the ocean basins. These recordings have also been used in the cloud-based collaborative analytics platform developed by CVision AI and annotated by experts from the University of Hawaii and OceansTurn.
Additionally, in 2010, the National Oceanic and Atmospheric Administration (NOAA) ocean exploration team aboard the NOAA ship Okeanos Explorer collected video data using a dual system of remotely operated vehicles. In order to annotate collected videos more extensively, they began funding professional taxonomists in 2015. Initially, they collectively obtained annotations through volunteer participating scientists. A portion of the MBARI dataset, as well as materials from National Geographic and NOAA, are included in FathomNet.
Since FathomNet is open source, other institutions can easily contribute and use it instead of more time- and resource-consuming conventional methods of processing and analyzing visual data. Additionally, MBARI initiated a pilot initiative to use machine learning models trained on FathomNet data to analyze video taken by remotely controlled underwater vehicles (ROVs). The use of AI algorithms increased the labeling rate tenfold and reduced human effort by 81 percent. FathomNet’s data-driven machine learning algorithms could revolutionize ocean exploration and monitoring. One such example includes the use of camera-equipped robotic vehicles and enhanced machine learning algorithms to automatically search for and track marine life and other things underwater.
With ongoing contributions, FathomNet currently has 84,454 images reflecting 175,875 locations from 81 different collections for 2,243 concepts. The dataset will soon have more than 200 million observations after obtaining 1,000 independent observations for more than 200,000 animal species at various positions and image setups. Four years ago, a lack of annotated photos prevented machine learning from examining thousands of hours of ocean film. However, by unlocking discoveries and enabling tools that explorers, scientists and the general public can use to accelerate the pace of ocean research, FathomNet makes this vision a reality.
FathomNet is a fantastic illustration of how collaboration and community science can foster innovations in our understanding of the ocean. The team believes the dataset can help accelerate ocean research when understanding the ocean is more crucial than ever, using data from MBARI and other collaborators as a foundation. The researchers also emphasize their desire for FathomNet to function as a community where ocean aficionados and explorers from all walks of life can share their knowledge and skills. This will act as a springboard to address issues with ocean visual data that would not otherwise have been possible without extensive participation. To speed visual data processing and create a healthy, sustainable ocean, FathomNet is constantly being improved to include more tagged data from the community.
This article is written as a research summary article by Marktechpost staff based on the research paper’FathomNet: A Global Picture
database to enable artificial intelligence in the ocean‘. All credit for this research goes to the researchers of this project. review the paper, tool and reference article. Also, don’t forget to join our 26k+ ML SubReddit, discord channel, and electronic newsletterwhere we share the latest AI research news, exciting AI projects, and more.
Khushboo Gupta is a consulting intern at MarktechPost. He is currently pursuing his B.Tech at the Indian Institute of Technology (IIT), Goa. She is passionate about the fields of machine learning, natural language processing, and web development. She likes to learn more about the technical field by participating in various challenges.
