In the UK, about 20% of cropland relies on insect pollination, for which the honey bee is the most important, so the 50% decline of the honey bee population since the second world war is of major concern.
The overall goal of the project is to explore the requirements for an integrated bee hive sensor that is connected to the internet and capable of recording various metrics relevant to the health of the bee colony. Remote sensing has the potential to provide instantaneous data that will allow best management of hives and contribute to the understanding of the factors that affect the health of the hive.
Since different types of sensors pose different challenges, in the initial iteration they are largely separate, though they share the same power supply–a solar panel and battery–and the same WiFi connection. Future iterations are likely to be more tightly integrated, and the data transmission minimized to allow for cellular connectivity for field deployments where WiFi is not available.
The video component of the sensor consists of two cameras–one on the inside of the hive facing a frame in the hive, and the other on the outside facing the hive entrance. Both cameras are RaspiCams, each connected to a Rasberry Pi Model 3B. The internal camera configuration includes infra-red LEDs that are controlled by the Raspberry Pi. For five minutes every hour, the Raspberry Pi turns on the LEDs (which are outside the visible spectrum for bees) and records video, which is then transferred over WiFi to a network storage device. The resolution of the video is quite high at 1920x1080 at a 30fps. The external camera relies on natural illumination; however, since bees are capable of moving extremely fast when they are flying, the frame rate is set to 79fps.
Once the video has been transferred, it is analyzed on a desktop, though future iterations may analyze the video on the Raspberry Pi to remove the necessity of transferring large video files. Part of the project is to determine what information we can gather from the video. For the internal hive video, an obvious starting point is to try and determine how many bees are visible in the frame, which may be a metric of hive population growth and decline. Using optical flow we can also determine the activity levels, and on how much of the frame that activity is taking place. Right now the bees are busy building comb, so later on it may be possible to analyze how much comb has been built. Since the video is of high quality, it may also be possible to analyze aspects of bee behavior; certain viral infections cause paralysis, which may be detectable.
With the external camera, one of the main goals is to determine how many bees are leaving and entering the hive, since this gives us a metric of the level of foraging behavior and should also be correlated with external factors, such as the weather. This requires tracking each bee from frame to frame to see where they go. Once this tracking is enabled, we can also look at other aspects of the behavior of bees in front of the hive; one observation is that at certain times the bees tend to hover in front of the hive. This is most likely the bee dance, and can be observed by calculating for how many frames a bee is tracked before entering the hive or leaving the area. While 79fps is probably not fast enough to examine details of the bee dance, at a lower resolution the camera has been able to record in excess of 120fps, so this is a potential for further exploration.
The advantage of an integrated sensor is that all this video data can be correlated with data from the other sensors, as well as the ground truth observations of the bee keepers. When there is enough data we can use machine learning to extract the predictors of hive health and the effects of hive management decisions.
Contact: rob@uw.edu
Collaborators: School of Computer Science, The University of Manchester
Funders: N/A