AUDRON: A Deep Learning Framework with Fused Acoustic Signatures for Drone Type Recognition

TL;DR

The paper addresses robust acoustic drone detection amid environmental noise by introducing AUDRON, a multi-branch deep-learning framework that processes raw audio through four parallel streams—MFCC with 1D-CNN, STFT-CNN, bidirectional LSTM with attention, and an autoencoder—whose features are fused for classification. Synthetic data generated from a rotor-harmonic model $x_c(t) = \left( \sum_{k=1}^{K_c} A_k \sin(2\pi f_k t) \right) \cdot M(t) + \eta(t) + \xi(t)$, along with real-world DroneAudioDataset augmented with noise classes, underpins robust learning. AUDRON achieves 98.51% binary and 97.11% multiclass accuracy, outperforming CNN, RNN, and CRNN baselines and demonstrating strong generalization with diverse data and environments. These results suggest AUDRON's potential for secure, real-time acoustic surveillance where vision or radar sensing may be limited.

Abstract

Unmanned aerial vehicles (UAVs), commonly known as drones, are increasingly used across diverse domains, including logistics, agriculture, surveillance, and defense. While these systems provide numerous benefits, their misuse raises safety and security concerns, making effective detection mechanisms essential. Acoustic sensing offers a low-cost and non-intrusive alternative to vision or radar-based detection, as drone propellers generate distinctive sound patterns. This study introduces AUDRON (AUdio-based Drone Recognition Network), a hybrid deep learning framework for drone sound detection, employing a combination of Mel-Frequency Cepstral Coefficients (MFCC), Short-Time Fourier Transform (STFT) spectrograms processed with convolutional neural networks (CNNs), recurrent layers for temporal modeling, and autoencoder-based representations. Feature-level fusion integrates complementary information before classification. Experimental evaluation demonstrates that AUDRON effectively differentiates drone acoustic signatures from background noise, achieving high accuracy while maintaining generalizability across varying conditions. AUDRON achieves 98.51 percent and 97.11 percent accuracy in binary and multiclass classification. The results highlight the advantage of combining multiple feature representations with deep learning for reliable acoustic drone detection, suggesting the framework's potential for deployment in security and surveillance applications where visual or radar sensing may be limited.

Figures (5)

• Figure 1: Flowchart of synthetic audio generation highlighting key components used in the process.
• Figure 2: Sample waveforms from the synthetic model, illustrating distinct audio signatures for the four drone classes.
• Figure 3: Sample waveforms (left) and spectrograms (right) from the experimental datasets.
• Figure 4: The complete architecture of our proposed multi-modal drone detection model, AUDRON. A raw audio waveform is processed by four parallel feature extraction branches: MFCC extractor, STFT-CNN, RNN, and Autoencoder. The resulting feature vectors are concatenated and passed through a fusion layer and final classification head to produce the output.
• Figure 5: Training dynamics and performance evaluation of the proposed model across different datasets and tasks. (a) Training and validation loss/accuracy curves for the model trained on synthetically generated data, showing rapid convergence. (b) Confusion matrix for binary classification on real-world data. (c) Confusion matrix for the multiclass classification task. (d) Confusion matrix for the model evaluated on the synthetic dataset. (e) Training and validation accuracy curves for the binary classification model. (f) Training and validation accuracy curves for the multiclass classification model.