Said Mustafa Sadat, “Fishing for Data: Using AI to Identify Fish Population Dynamics at Sleeping Bear Dunes in the Nearshore of Northeast Lake Michigan”
Mentor: Thomas Hansen, Freshwater Sciences
Poster #162

The Great Lakes is home to one of the largest ecosystems of freshwater in North America and management of the lakes is subject to several international treaties and regional compacts. Recently, several outbreaks of Avian Botulism in northeast nearshores of Lake Michigan have been under investigation. Collecting and interpreting underwater images and videos to assess the populations and movements of fish species in the area is a valuable tool used to identify and study the factors linked to these outbreaks. Manual data interpretation imposes both time and labor-intensive limitations in producing meaningful observations. To help address this issue, our laboratory is exploring automating the process of identifying fish. Leveraging the recent advancements of Deep learning, we identified YOLO-based Convolutional Neural Networks (CNN’s) as a potential approach. We seek to develop a framework which utilizes a YOLO-based model, mimicking the cognitive learning process of the human brain in identifying patterns. Two critical components of our project have been to preprocess collected imagery and to optimize our model’s performance. We aim to enhance our imagery by applying methods from the OpenCV library. To create training data, we designed a website to resemble a yes/no swiping app to enable the manual labeling of training images quickly and efficiently, using a familiar paradigm. By automating our data interpretation process through Deep Learning based YOLO models, we are eliminating the limitations of manual data interpretation. Automated AI-based decision making will allow us to draw a wider range of observations from existing and future data–in support of the effort to protect freshwater ecosystems from harmful phenomena.