Acoustic Analysis of Beehives for Precision Beekeeping Based on the Internet of Things: A Case Study on Improving Hive Health and Productivity

Document Type : Research Paper

Authors

1 Ph.D. Candidate, Department of Agricultural Machinery Engineering, Agriculture and Natural Resources Campus, University of Tehran, Tehran, Iran

2 Professor, Department of Agricultural Machinery Engineering, Agriculture and Natural Resources Campus, University of Tehran, Tehran, Iran

3 Associate Professor, Department of Mechanical engineering of biosystems, Razi University, Kermanshah, Iran

Abstract

     Pollination is essential for the sexual reproduction of many crops, fruits, and most wild plants. Among animal pollinators, solitary and social bees play a major role. In addition to their role in pollinating wild plants, managed honeybee colonies are economically the most valuable group of pollinators for monoculture crops and fruits worldwide. This study presents a novel method for diagnosing honeybee colony diseases and problems using sound analysis and deep learning. First, the sounds produced by the honeybee colony were recorded by designing a smart beehive and placing a microphone in an optimal location. Then, by converting the audio signals into spectrograms and using convolutional neural networks, sound patterns associated with various diseases and problems such as queen less ness, varroa mite infestation, and foulbrood and nosema diseases were identified. The results showed that this method is capable of diagnosing these problems with an accuracy of over 98%. For example, the model was able to detect queen less ness with 98.62% accuracy, the probability of varroa mite presence with 98.59% accuracy, and the probability of foulbrood disease with 98.71% accuracy. Finally, by implementing the Internet of Things in the hive management system, a significant improvement in the quantity and quality of honey produced was observed. This research shows that sound analysis and deep learning can be used as a powerful tool for monitoring the health of honeybee colonies and increasing productivity in the beekeeping industry.

Keywords

Main Subjects

EXTENDED ABSTRACT

Introduction

Pollination, achieved by insects, birds and other, is crucial for the sexual reproduction of numerous crops, fruits, and wildest plants. Among pollinators, insect pollination, particularly by solitary and social bees, plays a dominant role. Beyond their contribution to wild plant pollination, managed honey bees (primarily Apis mellifera) constitute the most economically valuable pollinator group in monocrops and fruits globally. The primary objective of this study was to develop a novel method for monitoring beehive health using sound analysis. By leveraging deep learning techniques, we aimed to accurately detect the presence or absence of a queen bee, as well as various diseases and pests within bee colonies.

Methodology

To achieve this, we initially designed a beehive and the optimal microphone placement within it using SOLIDWORKS 2022. Subsequently, we employed sound analysis to assess the health status of bee colonies. One-minute audio files were recorded from each hive and converted into spectrograms. These spectrograms were then analyzed using a Convolutional Deep Learning Neural Network (CDLNN). The performance of DenseNet121, EfficientNetB0, and InceptionV3 networks was generated and selected the best-performing one. The chosen network was further trained with six different combinations of batch size and epochs. Finally, spectrograms from healthy hives were compared with those from hives affected by various issues such as queen lessness, Varroa mites, tracheal mites, American/European foulbrood, and Nosema. This innovative system continuously records audio from beehives and analyzes the sounds to detect signs of illness. By using advanced techniques like the Fast Fourier Transform and deep learning, the system can identify various bee diseases with high accuracy. The collected audio data is processed on a cloud-based platform, where powerful artificial intelligence models analyze the sound patterns to identify any health issues within the bee colony. This technology provides beekeepers with a valuable tool to proactively monitor the health of their hives and take necessary steps to prevent colony collapse disorder.

GSMT represents the GSM transmitter modules and their associated circuitry, responsible for initiating calls to the GSM receiver. GSMR denotes the GSM receiver modules and their circuitry, designed to record audio transmitted from the beehives and store it. PC refers to the personal computer that serves as an interface between the GSM modules and Google Cloud. LAN signifies the local area network connecting the PC and the GSM receiver modules. Cloud Computing refers to the Google Cloud Platform, used for data storage and processing.

Ultimately, audio data is stored in Google Drive and processed on Google Colab using a deep learning system based on convolutional neural networks. If any of the defined problems occur (such as queen lessness, Varroa mites, tracheal mites, foulbrood, or Nosema disease), a message is sent to the beekeeper via Telegram. This message includes the audio file containing the detected disease, the processed spectrogram with the detection accuracy percentage, and a text message specifying the beehive number and the probability of the problem.

Results

Our findings demonstrated that sound analysis can accurately detect the presence or absence of a queen bee, as well as infestations by mites and diseases like foulbrood and Nosema. Each specific problem within a beehive produced a unique sound pattern that could be identified using neural networks. By analyzing the fast Fourier transform of one-minute audio recordings, spectrograms and extracted frequency features was generated to interpret the beehive's condition. The neural network achieved an accuracy of over 98% in identifying various issues (queen lessness 98.62%, Varroa mites 98.59%, tracheal mites 98.64%, American/European foulbrood 98.71%, and Nosema 98.80%). Moreover, the proposed Internet of Things system significantly improved the quality and quantity of honey produced compared to traditional beekeeping practices.

The neural networks achieved remarkably high accuracies in identifying various hive anomalies. Specifically, accuracies exceeding 98% were obtained for queenlessness (98.62%), Varroa mites (98.59%), tracheal mites (98.64%), American/European foulbrood (98.71%), and Nosema (98.80%).

Healthy hive with queen: Dominant frequencies were observed between 0 and 2000 Hz, indicative of a stable and healthy colony.

Queenless hive: Higher frequencies, ranging from 2000 to 4000 Hz, were detected, suggesting increased agitation and disorganization among the bees.

Varroa mite infestation: The dominant frequency range shifted to 1000-3000 Hz, indicating increased bee activity to combat the infestation.

Acariasis infestation: Lower frequencies, between 500 and 1500 Hz, were observed, suggesting decreased hive activity and a weakened colony.

American or European foulbrood: A significant decrease in the overall frequency range (400-1200 Hz) was noted, indicating a weakened colony and reduced bee activity.

Nosema infestation: Higher frequencies (800-2400 Hz) were observed, suggesting increased bee activity due to digestive system disturbances.

 These results were validated through comprehensive evaluation using confusion matrices and receiver operating characteristic (ROC) curves. The InceptionV3 architecture exhibited the most robust performance, achieving an overall accuracy of 98.74%.

Performance Optimization

To optimize the performance of the InceptionV3 model, hyperparameter tuning was conducted. By systematically varying parameters such as batch size and epochs, the optimal configuration for training was determined. The results indicated that a batch size of 16 and 20 epochs yielded the best performance, balancing accuracy and training time.

A deep learning-based analysis of beehive sounds, using Fast Fourier Transform (FFT), revealed distinct acoustic patterns associated with various hive health conditions. The following are the quantitative and qualitative impacts of the employed research method on honey production:

Enhanced Honey Yield: Hives managed with acoustic monitoring produced significantly higher honey yields (12 kg on average) compared to conventionally managed hives (8-10 kg).

Improved Honey Quality:

Increased Invertase Activity: The honey from acoustically monitored hives exhibited higher invertase activity, resulting in a greater proportion of simple sugars and a sweeter taste.

Reduced Sucrose Content: Lower sucrose levels indicated more complete conversion of nectar into honey.

Favorable Fructose-to-Glucose Ratio: A lower fructose-to-glucose ratio contributed to a lower glycemic index, making the honey suitable for individuals with diabetes.

Elevated Proline Content: Higher proline levels indicated better quality and a stronger flavor.

Increased Diastase Activity: Greater diastase activity pointed to higher enzymatic activity and improved shelf life.

Lower Moisture Content: Reduced moisture content contributed to better honey quality and longer shelf life.

Conclusion

The CDLNN-based IoT system introduced in this study accurately detected the issues under investigation (with an accuracy of over 98%), reducing manual intervention in beehive management and positively impacting honey quality and quantity. Consequently, this system can be widely recommended for beehive management to enhance honey quality and quantity, as well as provide non-invasive and easy monitoring for beekeepers.

Author Contributions

Payam Faramarzi (P.F.); Conceptualization, Methodology, Data collection, Data processing, Software, Validation, Writing—original draft preparation. Reza Alimardani (R.A.); Supervision, Project administration, Technical Supervision, Writing, review and editing. Hekmat Rabbani (H.R.); Review and editing, Formal analysis, Technical Consultation. Hossein Mousazadeh (H.M.); Review and editing, Formal analysis, Technical Consultation.

All authors have read and agreed to the published version of the manuscript.

Data Availability Statement

For data and results access, please contact the corresponding author via the following email address or ResearchGate page:

Payam.faramarzi@ut.ac.ir

https://www.researchgate.net/profile/Payam-Faramarzi-2

Ethical considerations

The authors avoided data fabrication, falsification, plagiarism, and misconduct.

Conflict of interest

The author declares no conflict of interest.

References

Aumann, H., Tautz, J., & Tscheulin, T. (2017). Measuring honeybee foraging activity with Doppler radar. Journal of Apicultural Research, 56(4), 417–425.
Barrionuevo, A. (2007, February 27). Honeybees vanish, leaving keepers in peril. The New York Times. Retrieved from https://www.nytimes.com/2007/02/27/science/27bees.html
Büchler, R. (2015). Varroa tolerance in honey bees—Occurrence, characters and breeding. Bee World, 96(2), 54–70.
De Jong, D., De Jong, P. H., & Gonçalves, L. S. (1982). Weight loss and other damage to developing worker honeybees from infestation with Varroa jacobsoni. Journal of Apicultural Research, 21(3), 165–167.  
  De Simone, A., Barbisan, L., Turvani, G., & Riente, F. (2024). Advancing beekeeping: IoT and TinyML for queen bee monitoring using audio signals. IEEE Transactions on Instrumentation and Measurement, 73, 2502214.
  Frisch, K. v. (1993). The dance language and orientation of bees (2nd ed.). Harvard University Press. (Original work published 1886)
  Gallai, N., Salles, J. M., Settele, J., & Vaissière, B. E. (2009). Economic valuation of the vulnerability of world agriculture confronted with pollinator decline. Ecological Economics, 68(3), 810–821.  
  Genersch, E. (2010). American foulbrood in honeybees and its causative agent, Paenibacillus larvae. Journal of Invertebrate Pathology, 103(Suppl. 1), S10–S19.  
  Iqbal, K., Alabdullah, B., Al Mudawi, N., Algarni, A., Jalal, A., & Park, J. (2024). Empirical analysis of honeybees acoustics as biosensors signals for swarm prediction in beehives. IEEE Access, 12, 115167-115181.
  Janetzky, P., Davidson, P., Steininger, M., Krause, A., & Hotho, A. (2021). Detecting presence of speech in acoustic data obtained from beehives. In Detection and Classification of Acoustic Scenes and Events 2021.
  Kim, J., Oh, J., & Heo, T. Y. (2021). Acoustic scene classification and visualization of beehive sounds using machine learning algorithms and Grad-CAM. Mathematical Problems in Engineering, 2021, Article 6671575.
  Martin, S. J. (1994). Ontogenesis of the mite Varroa jacobsoni Oud. in worker brood of the honeybee Apis mellifera L. under natural conditions. Experimental & Applied Acarology, 18(2), 87–100.  
  McMenamin, A. J., & Genersch, E. (2015). Honey bee colony losses and associated viruses. Current Opinion in Insect Science, 8, 121–129.  
  Morelle, R. (2013, March 27). Neonicotinoid pesticides ‘damage brains of bees’. BBC News. Retrieved from https://www.bbc.com/news/science-environment-21958547
  Morse, R. A., & Calderon, N. W. (2000). The value of honey bee pollination in the United States. Bee Culture, 128(1), 1–15.
  Nolasco, I., Terenzi, A., Cecchi, S., Orcioni, S., Bear, H. L., & Benetos, E. (2019). Audio-based identification of beehive states. In ICASSP 2019-2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP).  
  Polyniak, Y., Kretzschmar, M., & Sylvestre, J. P. (2019). Dynamic time warping for honeybee colony sound classification. Ecological Informatics, 50, 100983.
  Qandour, A., Ahmad, I., Habibi, D., & Leppard, M. (2014). Remote beehive monitoring using acoustic signals. Acoustics Australia, 42(3), 204–209.  
  Roberts, J. M. K., Anderson, D. L., & Durr, P. A. (2017). Absence of deformed wing virus and Varroa destructor in Australia provides unique perspectives on honeybee viral landscapes and colony losses. Scientific Reports, 7, Article 42824.  
  Robles-Guerrero, A., Gómez-Jiménez, S., Saucedo-Anaya, T., López-Betancur, D., Navarro-Solís, D., & Guerrero-Méndez, C. (2024). Convolutional neural networks for real time classification of beehive acoustic patterns on constrained devices. Sensors, 24(19), Article 6384.
  Seeley, T. D. (2010). Honeybee democracy. Princeton University Press.
  Sharif, M. Z., Wario, F., Di, N., Xue, R., & Liu, F. (2020). Soundscape indices: New features for classifying beehive audio samples. Sociobiology, 67(4), 566–571.  
  Wallner, K. (1999). Varroacides and their residues in bee products. Apidologie, 30(2-3), 235–248.
  Zhao, Y., Deng, G., Zhang, L., Di, N., Jiang, X., & Li, Z. (2021). Based investigate of beehive sound to detect air pollutants by machine learning. Ecological Informatics, 61, Article 101246.
  Zgank, A. (2021). IoT-based bee swarm activity acoustic classification using deep neural networks. Sensors, 21(3), Article 676.
Iranian Journal of Biosystems Engineering
Volume 55, Issue 3
October 2025
Pages 49-71

Files

History

  • Receive Date: 05 October 2024
  • Revise Date: 30 January 2025
  • Accept Date: 03 February 2025

Share

How to cite

Statistics