Remote sensing is an important tool for monitoring restoration activities associated with the energy industry. In particular stocking-density surveys are extremely time-intensive activities that can benefit from the development of new technologies. Drones and piloted aircraft acquire imagery that has the potential to assist, but locating and identifying young seedlings in complex scenes is not a trivial task. While a well-trained eye can see them, manual interpretation does not scale operationally over large areas. This poses the main motivation for the project, which is to use state-of-the-art machine learning techniques, in particular, Deep Learning, to effectively develop a tool that will learn from examples how to identify seedlings and therefore be instrumental in the assessment of a forest’s regeneration. In addition to assessing Deep Learning technology for performing automated stocking surveys, this work has the potential to inform about the overall regeneration process, e.g., to investigate whether there are co-location patterns among emerging species (a spatial data mining task) or location patterns associated with animal migration and the like.
This project contributes to the Boreal Ecosystem Recovery and Assessment (BERA) Project.
Michael Fromm (LMU), Matthias Schubert (LMU), Greg McDermid (UoC), Julia Linke (UoC), Guillermo Castilla (Canadian Forest Service, Man Fai Wu(UoC)