Bioacoustic Data Augmentation Techniques for Data-Driven Apiculture

This paper explores the potential of data augmentation techniques to improve the detection of bee activities using machine learning models applied to bioacoustic data. Traditional machine learning methods used in bee bioacoustics often face challenges due to the imbalance in recorded sounds, which can limit the effectiveness of these methods. To address this, we consider a variety of data augmentation techniques, including Generative Adversarial Networks and simpler methods like time-stretching, pitch-shifting, and noise addition. Although our primary focus is swarming detection, these methods are flexible and could be applied to other bee activity or health monitoring systems. While long- and short-term memory networks are used as an example, the proposed augmentation technique is versatile and can be utilized by any machine learning model applied to bee bioacoustics. Our results show significant improvements in model performance, including an increase in accuracy from 67% to 88%, demonstrating the value of data augmentation in enhancing bee bioacoustics analysis.