Non-Intrusive System for Honeybee Recognition Based on Audio Signals and Maximum Likelihood Classification by...

Area of Science:

• Bioacoustics
• Artificial Intelligence in Agriculture
• Entomology

Background:

• The integration of Artificial Intelligence (AI) and the Internet of Things (IoT) is revolutionizing beehive monitoring.
• Distinguishing between worker and drone bees is crucial for hive management and understanding colony health.
• Current methods for bee type identification may be labor-intensive or lack automation.

Purpose of the Study:

• To develop an automatic method for recognizing honeybee types (worker vs. drone) based on their flight sounds.
• To evaluate various audio signal preprocessing and representation techniques for accurate bee classification.
• To implement and validate an autoencoder neural network for discriminating bee types using acoustic features.

Main Methods:

• Collected and analyzed audio signals from worker and drone bees near a beehive entrance.
• Compared signal preprocessing techniques and frequency-domain representations: Mel-Frequency Cepstral Coefficients (MFCCs), Gammatone Cepstral Coefficients (GTCCs), MUSIC, and Burg's PSD estimation.
• Utilized an autoencoder neural network, classifying bees based on signal representation reconstruction error, employing novel thresholding strategies.

Main Results:

• Demonstrated that audio signal analysis alone is sufficient to differentiate between drone and worker bees.
• Identified effective signal preprocessing and representation methods for this acoustic classification task.
• Achieved a high level of detection accuracy, validating the proposed autoencoder-based approach.

Conclusions:

• The proposed method enables automatic and accurate differentiation of honeybee types using flight acoustics.
• This technology can form the basis for an efficient, automated system for beekeepers.
• Acoustic monitoring offers a non-invasive and scalable solution for bee colony management.