Abstract
Instrumentation systems such as the Adaptable Monitoring Package (AMP) have significantly advanced the state of the art in marine energy monitoring. However, high bandwidth instruments on these systems, such as optical cameras and imaging sonars, can produce large quantities of data when acquiring continuously, burdening researchers with the costs of storage and review. Periodic acquisition reduces the data burden, but risks missing rare events, such as marine animal interaction. Event detection is therefore necessary to automate collection of high bandwidth data. Building on the technologies and methodologies developed for the AMP, MarineSitu has made substantial progress in the use of object detection algorithms (e.g., YOLOv3) to detect events in camera and sonar data. The benefits of this approach are quantified through a review of results from several recent projects. Detector accuracy, data volume, and review time are compared across projects and against continuous and periodic approaches to acquisition.