The decline in global bee populations significantly threatens biodiversity and food production. Monitoring beehives are crucial for understanding colony health and behavior. In this paper, we propose an analysis of different beekeeping approaches by leveraging deep neural network techniques to analyze the buzzing activity within beehives across multiple greenhouse environments. Our focus is identifying the bee’s presence or absence through audio analysis. This is crucial for early detection of hive issues, optimizing hive management, and public awareness efforts to protect bees and ecosystem sustainability. The first step involves spectrogram analysis of the buzzing sounds recorded within the hives. This study employed three distinct deep learning models: DenseNet121, ResNet50, and InceptionV3. Experimental evaluation using the Open Source Beehive and NU-Hive datasets demonstrates the superiority of InceptionV3 with an accuracy of approximately 72%, outperforming the other models. We evaluated a distributed approach with the VGG16 model, achieving an accuracy of 65%. Our findings emphasize the importance of selecting appropriate deep learning architectures for bee species classification tasks. Moreover, the practical implications of our approach extend beyond identifying bees’ species, as it can be applied to monitor hive health, disease detection, and stress levels within colonies. Our work is in the direction of knowing the transition requirements between a centralized approach and federated learning with the support of the arrival of RedCap devices and 5G networks capable of offering ad-hoc slices for each service.

Spectrogram Based Bee Sound Analysis with DNNs: a step toward Federated Learning approach

Borgianni, Luca;Ahmed, Md Sabbir;Adami, Davide;Giordano, Stefano
2023-01-01

Abstract

The decline in global bee populations significantly threatens biodiversity and food production. Monitoring beehives are crucial for understanding colony health and behavior. In this paper, we propose an analysis of different beekeeping approaches by leveraging deep neural network techniques to analyze the buzzing activity within beehives across multiple greenhouse environments. Our focus is identifying the bee’s presence or absence through audio analysis. This is crucial for early detection of hive issues, optimizing hive management, and public awareness efforts to protect bees and ecosystem sustainability. The first step involves spectrogram analysis of the buzzing sounds recorded within the hives. This study employed three distinct deep learning models: DenseNet121, ResNet50, and InceptionV3. Experimental evaluation using the Open Source Beehive and NU-Hive datasets demonstrates the superiority of InceptionV3 with an accuracy of approximately 72%, outperforming the other models. We evaluated a distributed approach with the VGG16 model, achieving an accuracy of 65%. Our findings emphasize the importance of selecting appropriate deep learning architectures for bee species classification tasks. Moreover, the practical implications of our approach extend beyond identifying bees’ species, as it can be applied to monitor hive health, disease detection, and stress levels within colonies. Our work is in the direction of knowing the transition requirements between a centralized approach and federated learning with the support of the arrival of RedCap devices and 5G networks capable of offering ad-hoc slices for each service.
2023
979-8-3503-8254-9
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11568/1215234
 Attenzione

Attenzione! I dati visualizzati non sono stati sottoposti a validazione da parte dell'ateneo

Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 0
  • ???jsp.display-item.citation.isi??? ND
social impact