2023 Author: Bryan Walter | [email protected]. Last modified: 2023-05-21 22:24
Bee dance is a method of communication between bees in a hive, with the help of which they transmit information to each other about the location of a new food source. Recognition and decoding of bee dances, even despite the advent of high-speed cameras, until now have been limited by the need for manual processing, therefore they were not accurate enough and took a lot of time. German mathematicians have created an artificial intelligence system that can recognize and decipher the dances of bees with high accuracy. A preprint of the article was published at arXiv.org.
If a bee has discovered a new source of nectar, then after returning to the hive with the help of a special dance, it can transmit information about its location to the bees that are not currently employed: direction, distance to it and the amount of nectar available in it. For the first time, the dance of bees began to be studied by the German entomologist Karl von Frisch in 1920, who received the Nobel Prize for his discovery. The bee dance has two main stages: the waggle stage and the return stage. During the first stage, the body of the bee makes oscillatory movements from side to side, while moving forward in a straight line. In the second stage, the bee returns to the starting point of its movement. In this case, the return to the starting point of the dance occurs alternately from the right or left side, so that the entire trajectory of the bee's movement describes the figure eight. The average orientation of such a figure of eight with respect to the vertical gives information about the direction to the nectar source, and the duration of the dance tells about the distance to the source and the amount of nectar in it.
Deciphering bee dances is useful for both beekeepers and entomologists. Initially, it was carried out using a protractor and a stopwatch, now more accurate technologies based on the analysis of video recordings are used. However, the main stages of decryption have to be done manually. As a result, it takes a long time, and it is possible to analyze only a small number of bees in a short period of time, which leads to a noticeable decrease in accuracy.
In their new work, mathematicians from Berlin were able to fully automate the process of recognizing and decoding bee dances using an artificial intelligence-based system.
Diagram of the system for recording the movement of bees inside the hive
The neural network for analyzing video recordings received from cameras installed inside the hive consists of four main modules. The first module, using hierarchical clustering systems, registers the oscillatory movements of the bodies of bees in order to fix the moment when the dance begins. It is known that the vibrations of the body of a bee at the first stage of the dance occur with a constant frequency of about 13 Hertz. As soon as movements with the desired frequency are detected, the system detects the beginning of the dance, and its further movement is recorded as a dance. The number of various movements of bees inside the hive is huge, and dances with information about the location of food sources occupy only a small fraction of them. Therefore, the second module is used to filter the necessary data in the streaming video using a convolutional neural network. The neural network analyzes a set of 50 by 50 pixel images and determines whether the analyzed video actually contains a dancing bee. In their work, however, the scientists do not indicate what was used to train the neural network. The third module is responsible for determining the orientation of the dance of the bee relative to the vertical. And the fourth module, based on the data received from the first three, decodes the polar coordinates of the power source and determines its position.
Results of the analysis of bee dancing in a real hive. The white and green triangles mark the locations on the map of the hive and power source, respectively. Possible food source locations are shown in purple based on the analysis of individual bee dancing. The dotted line shows the averaged direction to the power source.
Scientists have tested the device for multiple dances and food sources. As a result, the expansion accuracy was 90 percent. The direction was determined to within 3 degrees.
Neural networks are often used to observe the behavior of animals and study the routes of their movement. So, recently, using a neural network, they developed an algorithm to scare away cats who want to mark the territory.
