WaveSpec-DeiT: A Wavelet-Inspired Spectral Transformer For RF-Based UAV Identification

Abstract

The rapid expansion of unmanned aerial vehicles (UAVs) has created significant security concerns for airports, military installations, and critical infrastructure. Unauthorized drone incursions near controlled airspace and sensitive facilities highlight the need for reliable non-cooperative detection systems. While radar, optical, and acoustic approaches can be effective in controlled settings, their performance often deteriorates under adverse weather, non-line-of-sight (NLoS), and cluttered operational environments. Passive radio frequency (RF) sensing provides a practical alternative by monitoring UAV communication emissions independent of visibility conditions. This paper proposes WaveSpec-DeiT, a transformer-based framework for UAV identification using RF spectrogram representations. This integrates wavelet-inspired multi-scale RF filtering with spectral-domain token modeling in a data-efficient transformer backbone. By capturing global time and frequency dependencies and spectral correlations, this enhances discrimination of subtle modulation patterns and hardware-induced signal characteristics, particularly under low signal-to-noise ratio (SNR) and multipath fading conditions. From a mobile data management (MDM) standpoint, the framework also supports streaming RF data ingestion, low-latency classification at edge sensing nodes, and the construction of indexable spectral embeddings for distributed UAV monitoring databases. Experimental evaluation demonstrates that WaveSpec-DeiT achieves an overall accuracy of 93.57%, outperforming baseline transformer configurations. WaveSpec-DeiT offers a scalable and deployable solution for real-time UAV monitoring, supporting airport perimeter protection and tactical battlefield airspace awareness where reliable low-altitude detection is operationally critical.

Department(s)

Electrical and Computer Engineering

Second Department

Computer Science

Comments

Army Research Laboratory, Grant W911NF-22-2-0208

Keywords and Phrases

mobile data management; passive drone detection; RF sensing and spectrogram; spectral attention; time and frequency analysis; transformer networks; UAV identification; wavelet-inspired filtering

International Standard Serial Number (ISSN)

1551-6245

Document Type

Article - Conference proceedings

Document Version

Citation

File Type

text

Language(s)

English

Rights

© 2026 Institute of Electrical and Electronics Engineers, All rights reserved.

Publication Date

01 Jan 2026

Link to Full Text

Share

 
COinS