DECODING BEE COLONY STATES: ANALYSIS OF FEATURE EXTRACTION METHODS IN ACOUSTIC SIGNATURE CLASSIFICATION FOR QUEENLESS DETECTION

Abstract

In the domain of apicultural colony management, the existence of queen in a beehive is indispensable for sustenance of the hive's overall health and productivity. Nonetheless, conventional methodologies for evaluating the status of the queen are labor-intensive and may intrude upon the colony. This research endeavors to explore the application of acoustic pattern classification to differentiate between queenright (the residence of a queen) and queenless (the non-residence of a queen) conditions within bee colonies. Through the monitoring and examining the sound signals pour out by bees, this work seek to discern distinct acoustic characteristics that are associated with each condition. Sophisticated algorithms in machine learning are utilized to effectively sort these sound patterns with remarkable precision, especially KNN and RF models with 94.7% and 93.6% mean accuracy respectively using traditional MFCC’s which is widely used feature extraction method. Further this work extended by employing Gabor-MFCC method, a new approach which is a hybrid method demonstrated improved accuracy and noise resilience by 2% to 3% across the models. Then GFCC and MFCC spectral fusion methods produce further improvements in accuracy.  The findings indicate that specific sound frequencies and patterns serve as reliable indicators of the queen's status, thereby offering a non-invasive and automated solution for apiarists. This methodology presents substantial prospects for the enhancement of hive management practices by facilitating real-time, continuous surveillance of queen status, ultimately contributing to the sustainability and productivity of apicultural operations.