Using machine learning in bee acoustic analysis to detect presence or absence of a queen bee in a Ugandan bee hive

Abstract

Bees inside a beehive produce different sounds in form of vibration and acoustic signals when responding to specific stimuli inside the hive. These vibroacoustic signals can vary with different states of the beehive for example, when the beehive is infested with pests, bees produce different vibroacoustic signals from when the queen bee dies. Experienced beekeepers can listen to these acoustic signals and be able to classify them in order to tell the state of the beehive for example when the bees are about to swarm, or when the queen bee dies so that they can intervene and act accordingly i.e. introduce a new queen to the hive. However, when the beekeeper is away, it becomes impossible for him/her to tell when a tragedy befalls the hive, and by the time they make their periodic inspection of the hive, it may be too late, with the bees having swarmed or the hive having died out completely. This project will employ machine learning, deep learning in particular to analyse the sound from the beehive to detect the presence or absence of the queen bee in real-time. The project was implemented with the ESP32 microcontroller onto which other hardware components were connected. Three deep learning neural networks: Artificial Neural Network, Convolutional neural Network and Recurrent Neural Network were compared and the audio data was first converted into Mel-Frequency Cepstral Coefficients (MFCC) features extracted using the Librosa Python library. The project will depend on the fact that the frequency components of acoustic signals from the hive vary with different states of the hive, to analyse the sound with machine learning and alert the beekeeper in real time with a notification/call to their phone using the SIM800L cellular module when the absence of the queen bee is detected. The project will also involve development of a web application that will offer an interface for the user to monitor the beehive state, allow the user classify their own audio sample recordings in any format, and have a community where different users can interact with other users about matters pertaining to beekeeping.