The Predator-Prey Encounter Detection (PrED) Model: Detecting Interactions Between Rotifers and Algae Using Computer Vision and Machine Learning.

Understanding species interactions is integral to understanding ecosystem function and predicting how disturbances may affect the services provided to society. Analysis of trophic interactions, however, requires a significant investment of time and resources. Using a simple predator-prey pairing of a semi-sessile rotifer and motile algae, the Predator-Prey Encounter Detection (PrED) model aims to streamline and semi-automate video analysis of species interactions using computer vision and machine learning techniques. Based on Ultralytics YOLOv5 architecture with an implementation of SORT tracking, the PrED model can estimate algal density and generate a "potential encounter" log with timestamps to allow researchers to skip through a video. After the model validation and testing steps, the PrED model achieved over 90% precision and recall scores. By reducing the time it takes to analyze a species interaction video, the PrED model will save researchers time and act as an efficient tool to study plankton dynamics.