Classification of audio samples by convolutional networks in audio beehive monitoring

Abstract

In the investigation, we consider the problem of classification of audio samples resulting from the audio beehive monitoring. Audio beehive monitoring is a key component of electronic beehive monitoring (EBM) that can potentially automate the identification of various stressors for honeybee colonies. We propose to use convolutional neural networks (ConvNets) and compare developed ConvNets in classifying audio samples from electronic beehive monitors deployed in live beehives. As a result, samples are placed in one of the three non-overlapping categories: bee buzzing (B), cricket chirping (C), and ambient noise (N). We show that ConvNets trained to classify raw audio samples perform slightly better than ConvNets trained to classify spectrogram images of audio samples. We demonstrate that ConvNets can successfully operate in situ on low voltage devices such as the credit card size raspberry pi computer.