Short-Time Fourier Transform for Detecting the Queen Bee State

This study explores the potential of sound analysis to detect the queenless in beehives, a significant challenge for beekeepers. Comparing Mel-Frequency Cepstral Coefficients (MFCCs) and Short-Time Fourier Transform (STFT) as feature extraction methods, our proposal shows that machine learning models utilizing low-frequency (below 1200 Hz) features extracted by STFT outperform these models using MFCCs derived from the full spectrum for the queen bee presence detection in beehives. The emphasis on low frequencies provides several benefits such as improved classification accuracy and enhanced computational efficiency by focusing on informative features. This paper contributes valuable insights for developing effective bee colony health monitoring systems.