Honey bees account for nearly $20 billion in US crop production, but the number of honey-producing bee colonies in the US has seen a decline since the 1950’s, due in part to poor nutrition among modern day colonies. In order to address this, we must understand how honey bee colony health is impacted by the diversity and abundance of flowering species within the 10 km range that bees may fly from their nests. Yet these efforts have been thwarted by the sheer size of the problem: manually characterizing the flowering plant species within such a large area can be an arduous task. We describe a pipeline for automatically mapping the species of flowering plants around a colony from drone imagery. Drones can capture finer detail than satellites but require that images be stitched together to create an orthomosaic that represents the entire landscape of interest. We developed two different strategies for stitching the images together, identifying plants within them, and classifying each plant by its species. The “default” strategy performs these classification on the orthomosaic, while the “experimental” strategy runs the classification step in parallel on the original drone images, instead.
Results/Conclusions
We evaluated these strategies by their speed, memory usage, and accuracy on a dataset containing California Buckwheat plants. We found substantial reductions of at least 89% in running time and memory usage using the experimental strategy. By classifying plant species on the original drone images rather than the orthomosaic, we were able to improve our pipeline’s classification accuracy from 89% to 94%. These improvements will be crucial if we hope to run the pipeline on larger landscapes, especially those large enough to encompass the 10 km range that honey bees may fly from their nests. We plan to use the pipeline to measure key factors known to influence honey bee foraging behavior, such as patchiness, scarcity, or patch quality. These could help assess how the density and heterogeneity of floral resources in agricultural landscapes may affect colony health.