Mapping plant species in forest ecosystems is crucial to advance our understanding of forest ecology, invasive species mitigation, carbon sequestration potential, and more. UAV-based surveys both have high resolution and are economically scalable, but are not often used in natural scenes due to matching ambiguity and articulated geometry (e.g. due to wind). The goal of this project is to research adapting ML-based species mapping approaches to use unprocessed UAV imagery. This project seeks to (1) develop approaches for sampling of informative views while (2) minimizing duplicate manual labeling effort and (3) computing geolocation by reconstructing geometry, with the end goal of (4) accurately labeling and categorizing plant species or health.

I’m participating in superUROP because I want to work on interesting research problems, and this is a great opportunity to learn from the experts in the field. As an AI major who has taken Advances in Computer Vision, I feel lucky to be able to use computer vision in my project on such an impactful topic as mapping forests.