Development of an Automated Monitoring Platform for Invasive Plants in a Rare Great Lakes Ecosystem Using Uncrewed Aerial Systems and Convolutional Neural Networks

We present a novel method for rapidly and precisely monitoring invasive plant species within rare and Great Lakes endemic coastal ecosystems. Our monitoring platform comprises: 1) an Uncrewed Aerial System capable of collecting high-resolution imagery in a precise and repeatable manner; 2) software enabling ecologists to annotate this imagery to identity invasive plant species of interest; 3) neural network–based algorithms for identifying targeted invasive plant species in the images; and 4) software for generating georeferenced density maps of invasive plant species infestations. We applied our monitoring platform to two lakeplain prairie remnants in southeastern Michigan and classifier performance was high for both invasive reed (Phragmites australis subsp. australis) and glossy buckthorn (Frangala alnus) (AUC values of 96.5% and 99.4%, respectively). We present invasive species density maps generated by deep Convolutional Neural Networks. These site-specific georeferenced maps quantify invasive plant species density and distribution and provide resource managers with actionable insight to gauge risk to the site, plan biodiversity restoration, and evaluate the efficacy of control efforts.