| 研究実績の概要 |
This project aimed to develop a fully-automatic bee-tracking system that would allow quantification of behavior without the need for marking individuals. We achieved this goal, developing a convolutional neural network that can identify the position and orientation of thousands of unmarked bees with an accuracy surpassing that of humans. We applied this system a whole field season of bee foraging, recording five colonies over multiple generations of workers. We are currently analyzing this massive data set, which includes data on foraging activity, weather, and should allow us to reconstruct how individual honey bee colonies allocate their workers in the face of environmental uncertainty. In addition, we are able to follow individual unmarked bees for several minutes at a time, reconstructing ~ 46% of the trajectories in 5-minute recordings and over 71% of the tracks for at least 2 minutes. These advances should enable fully automated behavioral studies of honey bee behavior, including those whose behavior has been altered either pharmacologically or with disease stressors.
|
