Senior ResearcherContact Olof
At RISE Learning Machines Seminar on November 30 2023, we have the pleasure to listen to Ben Weinstein, University of Florida, give his talk: General Models for Airborne Wildlife Detection.
Ecology is undergoing rapid transformation in data collection. The broad adoption of automated tools for ecological monitoring has created enormous datasets of unlabeled sensor data. To transform these raw data into ecological inference, we need automated approaches for finding, counting and describing organisms. Computer vision, a form of image-based artificial intelligence, is a highly effective tool for developing large models for image-based ecological modeling.
Using examples from airborne tree surveys and avian colony surveys, we demonstrate the benefits of airborne monitoring coupled with automated AI classification. Using multi-sensor deep learning models, we present landscape level maps for individual tree species for over 100 million canopy crowns within the NEON network. Using multi-temporal surveys from UAVs, we present an airborne workflow for estimating nesting phenology in Everglades nesting bird colonies.
While these results open new avenues for research at unprecedented scales, there remain several obstacles to bringing AI advances to all ecologists. The combination of technical skill, significant annotation effort, and lack of model sharing severely limit the adoption of airborne ecological monitoring. To move beyond highly customized models for every application, we need general models that serve as a baseline for common ecological tasks from airborne imaging. I discuss the potential for general models and the hope for data integration across acquisition systems, taxa and geography.
Ben Weinstein obtained his PhD in Ecology and Evolution from Stony Brook University in 2016. He is currently a postdoctoral fellow at University of Florida, and studies the mechanisms that generate, maintain, and preserve biodiversity. Specifically, this includes the ecology of species interactions, hierarchical Bayesian models for ecological predictions, and machine learning for biodiversity image analysis.