QueenBuzz: A CNN-based architecture for Sound Processing of Queenless Beehive Towards European Apis Mellifera Bee Colonies' Survivability

Abstract

Honeybee colonies missing their queens are more likely to swarm and experience a fall in population. Bee growers in the Philippines still utilize traditional methods to determine the health of a hive. Traditional methods lead to difficulties if a hive goes without a queen for an extended period. The study focuses on how sound data may be used as input to a CNN-based architecture to determine whether a beehive has a queen. The research involves preparing audio files for conversion into a spectrogram, converting audio data into a spectrogram, converting the spectrogram into a Mel frequency cepstral coefficient, constructing and training a model for a feature based on the features of the spectrogram that is provided, and, as the last step, assessing the model with audio files that are different from the data used in the study. The study employs four CNN-based architectures for the training and evaluating of the model containing audio recordings taken from various beehives, each of which either lacked a queen or had one present. The simplified CNN model has an accuracy of 99.88% when predicting the sound of a queen-right hive, and it has an accuracy of 99.72% when predicting the sound of a queen-less hive.