DeepBees - Building and Scaling Convolutional Neuronal Nets For Fast and Large-Scale Visual Monitoring of Bee Hives

Julian Marstaller, Frederic Tausch, Simon Stock; Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2019, pp. 0-0

The decline of bee populations is a global trend and a severe threat to the ecosystem as well as to pollinator-dependent industries. Factor analysis and preventive measures are based on snapshot information. Information about the health state of a hive is infrequently acquired and remains labor-intensive and costly. In this paper, we describe a system that enables near-time, scalable, and cost-efficient monitoring of beehives using computer vision and deep learning. The systems pipeline consists of four major components. First, hardware at the hive gate is capturing the in and out streams of bees. Secondly, an on-edge inference for bee localization and tracking of single entities. Thirdly, a cloud infrastructure for device and data management with near-time sampling from devices. Fourthly, a cloud-hosted deep convolutional neuronal net inferring entity-based health insights. This MultiNet architecture, which we named DeepBees, is the main focus of this paper. We describe the development of the architecture and the acquisition of training data. The overall system is currently deployed by apic.ai and monitors 49 beehives in Karlsruhe in the south of Germany.