Using Deep Learning to Count Monarch Butterflies in Dense Clusters

Abstract

Monarch butterflies display one of the most fascinating migration patterns of all species, traveling over 3000 miles from their North American breeding grounds to reach overwintering sites in Central Mexico. Recent studies have suggested that monarchs have experienced an alarming decline in population size due to a combination of deforestation, loss of native milkweed and nectaring plants, and climate change. An issue that conservation efforts face is the lack of principled mechanisms to accurately estimate and count the population size of monarchs. This difficulty occurs due to their small size and existence in dense overwintering clusters in forests. We create an open-source tool to aid conservationists estimate the count of monarch butterflies from images automatically. To the best of our knowledge, our approach, based on deep convolutional neural networks, is the first automated application that can count small insects like monarch butterflies in dense clusters. We demonstrate that our approach achieves high accuracy in counting the number of butterflies even in the presence of occlusion. We also release an open-source dataset containing high resolution images of monarch butterflies along with human annotations for each butterflys position. Our open-source implementation can be readily used by scientists to estimate monarch numbers in overwintering clusters and could also be adapted for use in other clustering species.