A newly proposed acoustic detection framework based on rational Gaussian wavelet neural networks successfully detects different types of unmanned aerial vehicles with over 90 percent accuracy, maintaining this performance even within noisy operational settings. The system is designed to provide automated classification and detection using sound signals.
The development addresses the monitoring and detection of unmanned aerial vehicles, which represents an important capability for safeguarding civilian infrastructure as well as military infrastructure. To establish a cost-effective alternative to other detection approaches, the proposed system operates directly on acoustic signals obtained from microphones.
Once recorded, the incoming sound signals are directed through an integrated signal processing pipeline. Within this pipeline, the architecture employs interpretable adaptive feature extractors based on rational Gaussian wavelets to process the recorded acoustic data and capture key signal features.
These adaptive wavelet transformations are directly embedded into an underlying small neural network and trained jointly alongside it. By coupling the feature extraction pipeline with the neural network, the system detects and classifies unmanned aerial vehicles based on the obtained features, resulting in a physically interpretable machine learning algorithm.
In addition to identifying and classifying individual unmanned aerial vehicles, the physically interpretable system is capable of detecting drone swarms. The researchers demonstrated their results using empirical data collected across two distinct acoustic settings, evaluating performance in an indoor studio alongside noisy outdoor environments.
Based on experimental evaluations, the authors conclude that the proposed method outperforms traditional machine learning approaches for detecting and classifying single aerial vehicles as well as groups of multiple drones. While achieving this higher performance, the model maintains a high degree of interpretability, and the authors made the implementation of the proposed methods publicly available for reproducibility.
Subject of Research: Acoustic detection and classification of unmanned aerial vehicles
Article Title: Classification and Detection of Multiple UAVs using Rational Gaussian Wavelet Neural Networks
Article References: Ungvári, G., Braun, F., Ámon, A., Kackstädter, P., Volk, J., Kovács, P., & Dózsa, T. (2026). Classification and Detection of Multiple UAVs using Rational Gaussian Wavelet Neural Networks. Journal of Intelligent & Robotic Systems. https://doi.org/10.1007/s10846-026-02453-2
Image Credits: AI Generated
DOI: 10.1007/s10846-026-02453-2
Keywords: unmanned aerial vehicles, UAV detection, rational Gaussian wavelets, wavelet neural networks, acoustic signal processing, microphone sensors, drone swarm detection, infrastructure protection, machine learning classification, physically interpretable AI, audio feature extraction, indoor studio acoustic testing, noisy outdoor environments, acoustic surveillance, drone classification, adaptive wavelets, robotic systems
News Source: Cassandra Pierce. (November 20, 2026). Acoustic Wavelet Neural Network Detects Multiple Drones with Over 90 Percent Accuracy. Scienmag.



