AI-enhanced bee bio-acoustics with data augmentation and applications in early swarming detection
Loading...
Supplementary material
Other Title
Authors
Dangwal, Kartikeye
Author ORCID Profiles (clickable)
Degree
Master of Applied Technologies (Computing)
Grantor
Unitec, Te Pūkenga - New Zealand Institute of Skills and Technology
Date
2024
Supervisors
Ardekani, Iman
Varastehpour, Soheil
Varastehpour, Soheil
Type
Masters Thesis
Ngā Upoko Tukutuku (Māori subject headings)
Keyword
New Zealand
honeybees (Apis mellifera)
swarming detection
audio data processing systems
beehive monitoring
Artificial Neural Network (ANN)
Long Short-Term Memory (LSTM)
Generative Adversarial Networks (GANs)
honeybees (Apis mellifera)
swarming detection
audio data processing systems
beehive monitoring
Artificial Neural Network (ANN)
Long Short-Term Memory (LSTM)
Generative Adversarial Networks (GANs)
ANZSRC Field of Research Code (2020)
Citation
Dangwal, K. (2024). AI-enhanced bee bio-acoustics with data augmentation and applications in early swarming detection (Unpublished document submitted in partial fulfilment of the requirements for the degree of Master of Applied Technologies (Computing)). Unitec, Te Pūkenga - New Zealand Institute of Skills and Technology https://hdl.handle.net/10652/6501
Abstract
This thesis explores the application of advanced data augmentation and machine learning techniques to improve swarming detection through bee bio-acoustics. Traditional methods of beekeeping and their global practices, along with the history and importance of beekeeping in New Zealand, are discussed. The study addresses significant challenges in beekeeping, including environmental factors, pests, diseases, and data imbalance in audio samples. By leveraging Generative Adversarial Networks (GANs) and Long Short-Term Memory (LSTM) networks, we aim to enhance the detection of piping sounds, which are crucial for monitoring the health of bee colonies. The methodology includes collecting and preprocessing audio data, applying noise reduction, segmenting audio files, extracting features, and creating annotated data frames. Our experiments demonstrate that augmenting the original dataset with generated and denoised piping audio segments significantly improves model performance. The LSTM model's accuracy increased from 67% with the original dataset to 88% with the augmented dataset. Comparative analysis of precision, recall, and F1-scores further validated the effectiveness of our approach. Despite the successes, future work is needed to refine the generated audio samples to more closely resemble original sounds, and to develop more robust GAN architectures. This study's proposed methodology can be broadly applied to any audio dataset with imbalanced classes, offering a versatile solution for enhancing machine learning models in various domains.
Publisher
Permanent link
Link to ePress publication
DOI
Copyright holder
Author
Copyright notice
All rights reserved