New AI Framework Maps Invasive Mussels in Canadian Lakes
Researchers at the Vector Institute for Artificial Intelligence have developed a novel computer vision framework to help biologists map, monitor, and manage invasive zebra and quagga mussels in Canada’s freshwater lakes. Led by PhD student Angus Galloway and Research Director Graham Taylor, the project utilizes machine learning to automatically identify individual mussels in underwater imagery, determining coverage, abundance, and biomass. This innovation addresses the limitations of manual data collection by scuba divers, which is often hazardous, time-consuming, and limited in scope. The framework demonstrated significant accuracy, achieving 85%, 79%, and 71% agreement with expert human analysis for coverage, abundance, and biomass, respectively. By processing images in milliseconds, the tool enhances cost-effectiveness and multiplies biologist productivity, allowing for precise, up-to-date density maps crucial for protecting ecosystems and infrastructure like water intakes. Initially trained on 1,600 photos from Lake Erie and Lake Ontario, the open-sourced model offers potential for broader environmental applications, including detecting other invasive species like the round goby fish. This development represents a significant step forward in using artificial intelligence for aquatic ecology management, enabling targeted treatments and reducing the harmful impacts of invasive species.
Wire timeline
New AI Framework Maps Invasive Mussels in Canadian Lakes
Researchers at the Vector Institute for Artificial Intelligence have developed a novel computer vision framework to help biologists map, monitor, and manage invasive zebra and quagga mussels in Canada’s freshwater lakes. Led by PhD student Angus Galloway and Research Director Graham Taylor, the project utilizes machine learning to automatically identify individual mussels in underwater imagery, determining coverage, abundance, and biomass. This innovation addresses the limitations of manual data collection by scuba divers, which is often hazardous, time-consuming, and limited in scope. The framework demonstrated significant accuracy, achieving 85%, 79%, and 71% agreement with expert human analysis for coverage, abundance, and biomass, respectively. By processing images in milliseconds, the tool enhances cost-effectiveness and multiplies biologist productivity, allowing for precise, up-to-date density maps crucial for protecting ecosystems and infrastructure like water intakes. Initially trained on 1,600 photos from Lake Erie and Lake Ontario, the open-sourced model offers potential for broader environmental applications, including detecting other invasive species like the round goby fish. This development represents a significant step forward in using artificial intelligence for aquatic ecology management, enabling targeted treatments and reducing the harmful impacts of invasive species.
Vector Institute for Artificial Intelligence