Spectral Components of Honey Bee Sound Signals Recorded Inside and Outside the Beehive: An Explainable Machine...

Area of Science:

• Animal Behavior
• Bioacoustics
• Machine Learning for Ecology

Background:

• Honey bee behavior exhibits diurnal patterns critical for colony health.
• Automated monitoring systems require accurate time-of-day classification for activity and anomaly detection.
• Understanding acoustic signatures is key to non-invasive bee monitoring.

Purpose of the Study:

• To evaluate the effect of microphone placement on honey bee audio monitoring.
• To establish a baseline method for recognizing diurnal bee activity patterns.
• To compare the performance of different machine learning classifiers based on recording location.

Main Methods:

• Simultaneous audio acquisition from internal (brood frame) and external hive locations.
• Sound signal preprocessing using Power Spectral Density (PSD).
• Training Extra Trees and Convolutional Neural Network (CNN) classifiers for diurnal pattern recognition.

Main Results:

• Extra Trees classifier achieved 98-99% accuracy with internal recordings but only 61-72% with external recordings.
• CNNs achieved 76-87% accuracy with external recordings, outperforming Extra Trees in this setup.
• Analysis identified the 100-600 Hz frequency band, with peaks at 100 Hz and 300 Hz, as important for bee activity classification.

Conclusions:

• Internal hive microphone placement is superior for accurate honey bee time-of-day classification.
• The 100-600 Hz frequency band is critical for monitoring bee activity.
• Future monitoring systems should prioritize internal hive acoustics for validation and improved accuracy.