A.I. Sukachev1, R.S. Domrachev2, E.A. Sukacheva3

1-3 Voronezh State Technical University (Voronezh, Russia)

1 mag.dip@yandex.ru; 2 ggromagg@list.ru; 3 elena_s_1331@mail.ru

Problem statement. The proliferation of unmanned aerial vehicles (UAVs) requires the creation of reliable detection systems. Existing methods (radar, optical) have significant disadvantages: high cost, dependence on weather conditions and visibility, as well as difficulties in detecting small-sized drones against a complex background. Acoustic detection is a promising passive technology independent of these limitations. However, the development of accurate acoustic models faces a number of fundamental challenges. The key ones are the chronic shortage of labeled UAV sound data, the presence of intense background noises of various nature, as well as the effect of class imbalance in the training data. This imbalance, when there are significantly fewer examples with drones than background recordings, leads to a shift in algorithms towards the majority class and a sharp decrease in their effectiveness in detecting precisely targeted objects.

Goal. Development and experimental verification of a neural network approach for detecting UAVs by acoustic signal based on transfer learning, which provides high accuracy in conditions of limited data and background noise.

Results. An approach based on transfer learning with a pre-trained YAMNet model is proposed. Data augmentation and weighted loss function are used to solve the class imbalance problem. Hyperparameter optimization is performed using the grid search method. Experiments have shown stable growth of the F1-measure for the drone class, which confirms the robustness of the model to real noise. The developed solution is intended for integration into security systems as a reliable passive detection channel, especially in conditions of limited visibility.

Practical significance. The results of the study can be used in the design of UAV detection systems using the acoustic method.

Sukachev A.I., Domrachev R.S., Sukacheva E.A. Neural network approach to detecting acoustic signals of unmanned aerial vehicles based on transfer learning. Radiotekhnika. 2026. V. 90. № 4. P. 84−94. DOI: https://doi.org/10.18127/j00338486-202604-11 (In Russian)

1. Martynyuk V.D., Rybnikov D.A., Sukachev A.I., Sukacheva E.A. Programmno-apparatnyj kompleks obnaruzheniya i identi-fikacii bespilotnyh letatel'nyh apparatov. Vestnik Voronezhskogo gos. tekhnich. un-ta. 2024. T. 20. № 3. S. 97-102. DOI: 10.36622/1729-6501.2024.20.3.014 (in Russian).
2. Basak H. et al. Development of an acoustic system for UAV detection. Sensors. 2020. V. 20. № 17. P. 5015. URL: https://doi.org/10.3390/s20175015.
3. Kundu S. et al. Acoustic based drone detection via machine learning. IEEE Access. 2022. V. 10. P. 9989229. URL: https://doi.org/10.1109/ACCESS.2022.9989229.
4. Transfer learning with YAMNet for environmental sound classification. TensorFlow. URL: https://www.tensorflow.org/tuto-rials/audio/transfer_learning_audio (data obrashcheniya: 01.10.2025).
5. Sound classification with YAMNet. TensorFlow Hub. URL: https://www.tensorflow.org/hub/tutorials/yamnet (data obrashcheniya: 01.10.2025).
6. Сhollet F. Deep Learning with Python. 2nd Edition. Manning Publications. 2021. 510 p.
7. Classification: Accuracy, recall, precision, and related metrics. Google Developers. URL: https://developers.google.com/machine-learning/crash-course/classification/accuracy-precision-recall (data obrashcheniya: 02.10.2025).
8. Gemmeke J.F. Audio set: an ontology and human-labeled dataset for audio events. IEEE ICASSP. 2017. P. 776-780. URL: https://doi.org/10.1109/ICASSP.2017.7952261.
9. Howard A.G. et al. MobileNets: efficient convolutional neural networks for mobile vision applications. arXiv. 2017. arXiv:1704.04861. URL: https://arxiv.org/abs/1704.04861.
10. Roberts L. Understanding the mel spectrogram. Medium, Analytics Vidhya Section. 2020. URL: https://medium.com/analytics-vidhya/understanding-the-mel-spectrogram-fca2afa2ce53 (data obrashcheniya: 30.09.2025).
11. McEdward. Data augmentation for audio. Medium Blog. URL: https://medium.com/@makcedward/data-augmentation-for-audio-76912b01fdf6 (data obrashcheniya: 30.09.2025).
12. Pedregosa F. et al. Scikit-learn: machine learning in Python. Journal of Machine Learning Research. 2011. V. 12. P. 2825-2830.
13. How to improve class imbalance using class weights in ML ?. Analytics Vidhya. 2020. URL: https://www.analytic-svidhya.com/blog/2020/10/improve-class-imbalance-class-weights/ (data obrashcheniya: 01.10.2025).