Machine Learning Framework for the Acoustic Detection of the Queen Bee Presence - ELETTRONICO - conference paper

Aug 3, 2026

Upload by Noah

Honeybees are essential pollinators for crop production and biodiversity, yet rising mortality has increased the need for continuous, non-invasive monitoring. To support research and beekeepers, a machine learning framework is developed to detect the queen bee’s presence from bee-produced sound. The study applies Short Time Fourier Transform and Mel Frequency Cepstral Coefficient audio features with support vector machines and neural network classifiers. Results demonstrate the promise of machine learning to classify acoustic events linked to colony health.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages5

Reading Time13 min

AccessFree to download

Introduction

Developed Framework

Dataset preparation and audio chunking

Feature extraction and train-test split

Training with k-fold cross-validation

Evaluation and metrics logging

POLITECNICO DI TORINO Repository ISTITUZIONALE Machine Learning Framework for the Acoustic Detection of the Queen Bee Presence Original Machine Learning Framework for the Acoustic Detection of the Queen Bee Presence / Barbisan, Luca; Riente, Fabrizio. -ELETTRONICO. - (2024), pp. 4347-4350. (Intervento presentato al convegno Forum Acusticum 2023 tenutosi a Torino (Italy) nel 11-15 Settembre 2023) [10 .61782/fa.2023. 1309] . Availability: This version is available at: 11583/2991401 since: 2024-07-31T22:01:38Z Publisher: European Acoustics Association 2023 Published DOI:10.61782/fa.2023.1309 Terms of use: This article is made available under terms and conditions as specified in the corresponding bibliographic description in the repository Publisher copyright (Article begins on next page) 18 September 2024 DOI: 10.61782/fa.2023.1309 MACHINE LEARNING FRAMEWORK FOR THE ACOUSTIC DETECTION OF THE QUEEN BEE PRESENCE Luca Barbisan Dept. of Electronics and Telecommunications Politecnico di Torino, Torino, 10129 Italy luca.barbisan@polito.it Fabrizio Riente Dept. of Electronics and Telecommunications Politecnico di Torino, Torino, 10129 Italy fabrizio.riente@polito.it ABSTRACT Honeybees are one of the most important pollinators in nature for both crop production and biodiversity preservation. The increase in bee mortality observed in the last decade motivated the development of continuous monitoring systems to better understand this phenomenon. Different solutions have been presented in the literature, and particularly sound analysis appears the most promising among the non-invasive techniques. In this context, we developed a machine learning framework for the analysis of the sound produced by bees for the detection of the queen bee’s presence. The presence of the queen is an important indicator of the colony’s health. In this work, we investigated Short Time Fourier Transform and Mel Frequency Cepstral Coefficient audio features with support vector machines and neural network classifiers. The results indicate the potential of machine learning methods for supporting the researchers’ study and beekeepers in managing such important insects. Keywords: sound classification, machine learning, queen bee detection, bee monitoring. 1. INTRODUCTION The importance of honey bees (Apis mellifera L.) is not limited to the production of honey, beeswax, royal jelly, and propolis, but they provide pollination service for about 70% of the crop in the world. In the last decade, many stress factors such as climate change, the use of pesticides, and intensive agriculture have led to a decline in the honeybee colonies [1] . This situation emphasized the necessity of continuous monitoring to investigate the motivation for such a decline in the bee colonies and support both researchers and beekeepers. In the literature, several techniques have been proposed to tackle this problem. Some solutions collect environmental parameters inside the beehive [2–4] . Others, exploit computer vision to track bees’ movements [5, 6] . In some cases, the proposed solutions require the modification of the beehive for the installation of sensors, which is impractical for a real device. In recent years, non-invasive techniques, based on audio processing, are becoming interesting, thanks also to the miniaturization of sensing technologies. Honeybees communicate using a combination of vibroacoustic signals [7, 8] . The sound level amplitude and frequency depend on the activity of the colony. Frequencies can reach up to 3000 Hz in case of defensive reaction [9][10] . Researchers have proved the correlation between signal amplitudes and frequencies of honeybees and some events such as swarming [9,11] and queen presence [12,13] . In this scenario, the representation of acoustic signals, combined with machine learning methods can lead to the development of automatic systems that can discriminate among different events that can char

FAQs

Why is detecting the queen bee’s presence important?

The queen’s presence serves as an important indicator of the colony’s health, helping assess and manage the hive state.

What audio features and classifiers are used in the framework?

The framework investigates STFT and MFCC features combined with support vector machines and neural network classifiers.

How is model performance evaluated?

The approach trains models using k-fold cross-validation and performs final evaluation by extracting a log file containing the computed metrics.

Aug 3, 2026

Upload by Noah