Ocean Seafloor Monitoring
Scient harnesses the latest technology in artificial intelligence to classify species and substrate types through data-driven analytics. Our data analytics platform classifies habitats, coastal environments, and other natural resources such as gas hydrates in a non-invasive and non-destructive way.
Scient customizes and designs AI-based data processing workflows to tackle big data challenges in ocean conservation.
Our research strives to expand AI for ocean conservation. Using deep learning, the identification of a target species could be performed the same way an autonomous car can detect the presence of obstructions. To a properly trained deep convolutional neural network, layers of visual attributes are created to identify and distinguish biotic and abiotic features. Our first prototype incorporated this multi-layer neural network to indicate the presence or absence of 5 common species and substrates using training data. We significantly improved the taxonomic resolution of seafloor characteristics in our current prototype by measuring the abundance of greater than 20 classes of coastal species and substrates.
Scient offers an unprecedented spatial resolution for a non-invasive and non-destructive way of monitoring coastal environments in a fraction of time compared to traditional methods that rely on human expertise.
We appreciate the value of previously processed legacy datasets to train our deep neural networks before implementing them to new data.
Our service reduces the time required to analyze imagery data, which has been historically time and resource consuming, by three orders of magnitude.
Scient stands out from its competitor because it is a fully automated and AI-powered workflow.
Scient’s current functionality is capable of quantifying the abundance of species and substrate types in imagery data with the goal to aid in long-term monitoring in ocean environments. Once established, Scient can offer a transferrable, standardized protocol to integrate its state-of-the-art image analyzing technology in marine conservation and exploration activities.
Current State of Affairs:
Globally, Marine Protected Areas cover 7.68 % of the ocean, including National Waters (EEZ), and Areas Beyond National Jurisdiction. By 2030, many countries have signed an international agreement to designate 30 % of their EEZ as MPAs. With only 17.79 % of EEZ currently protected, there remains plenty of work to reach this goal.
Advanced marine exploration techniques such as autonomous underwater vehicles (AUVs) and remotely operated vehicles (ROVs) enable marine imaging and habitat monitoring on improved spatial and temporal scales. Traditionally, a comprehensive interpretation of ocean imagery data collections is performed by domain experts. This task takes on average 15-20 minutes of specialized annotation for each image, covering 1.5-10 square meters.
The Canadian government has budgeted 1 billion dollars to protect ocean and fisheries over the next five years. However, monitoring 30 % of Canada’s EEZ (1.7 million km2) over a period of five years would require 200,000 experts working full time at minimum wage, and cost 4.5 billion dollars. We believe that Scient can reduce the required time and budget for analyzing the ocean imagery data by three orders of magnitude.