RFUAV: A Benchmark Dataset for Unmanned Aerial Vehicle Detection and Identification (2503.09033v2)

Abstract: In this paper, we propose RFUAV as a new benchmark dataset for radio-frequency based (RF-based) unmanned aerial vehicle (UAV) identification and address the following challenges: Firstly, many existing datasets feature a restricted variety of drone types and insufficient volumes of raw data, which fail to meet the demands of practical applications. Secondly, existing datasets often lack raw data covering a broad range of signal-to-noise ratios (SNR), or do not provide tools for transforming raw data to different SNR levels. This limitation undermines the validity of model training and evaluation. Lastly, many existing datasets do not offer open-access evaluation tools, leading to a lack of unified evaluation standards in current research within this field. RFUAV comprises approximately 1.3 TB of raw frequency data collected from 37 distinct UAVs using the Universal Software Radio Peripheral (USRP) device in real-world environments. Through in-depth analysis of the RF data in RFUAV, we define a drone feature sequence called RF drone fingerprint, which aids in distinguishing drone signals. In addition to the dataset, RFUAV provides a baseline preprocessing method and model evaluation tools. Rigorous experiments demonstrate that these preprocessing methods achieve state-of-the-art (SOTA) performance using the provided evaluation tools. The RFUAV dataset and baseline implementation are publicly available at https://github.com/kitoweeknd/RFUAV/.

Analysis of RFUAV: A Benchmark Dataset for UAV Detection and Identification

The paper "RFUAV: A Benchmark Dataset for Unmanned Aerial Vehicle Detection and Identification" introduces a comprehensive benchmark dataset specifically designed for radio-frequency (RF)-based unmanned aerial vehicle (UAV) detection and identification. This research addresses several critical challenges in the domain, including the limited variety of drone types in existing datasets, the need for data across diverse signal-to-noise ratios (SNRs), and the lack of standardized evaluation tools.

Dataset Characteristics and Methodology

RFUAV comprises approximately 1.3 TB of raw frequency data collected from 37 different UAVs using the Universal Software Radio Peripheral (USRP) device in real-world environments. Unlike prior datasets like DroneRF and DroneRFa, which either limited to a few types or did not account adequately for SNR levels, RFUAV provides a wide spectrum of SNRs and tools to transform raw data across these levels. The dataset is noted for its meticulous coverage, including data from diverse operational scenarios of each UAV type, ensuring a broad representation of signal characteristics.

Distinct Features and Signal Analysis

A notable contribution of this dataset is the introduction of the "RF drone fingerprint," which captures unique spectral signatures of drones. This fingerprint is derived from analyzing key signal attributes, such as frequency-hopping spread spectrum (FHSS) characteristics, video transmission signal bandwidths, and other unique properties. Such detailed analysis enables more accurate differentiation of drone types based on RF signals.

The paper critically assesses how drone signals, often plagued by interference from other sources such as WiFi and Bluetooth, can be effectively distinguished. The inherent structure in FHSS signals—characterized by fixed bandwidths, hopping durations, and patterns—alongside temporal fluctuations in video transmission signals, forms the basis of their differentiation method.

Two-Stage Drone Detection and Identification Model

To harness the RFUAV dataset, the researchers developed a two-stage model combining object detection and image classification networks. The first stage employs YOLO for detecting drone signals in the time-frequency spectrogram, while the second stage utilizes ResNet for identifying drone types. This model demonstrates high efficiency and accuracy, especially under varying SNR conditions, thus affirming the robustness and practicality of RFUAV in real-world applications.

Experimentation and Findings

The paper includes extensive experimentation on the dataset, revealing insights into how different signal preprocessing parameters and model architectures influence detection outcomes. Experiments demonstrate that specific preprocessing pipelines, such as choice of color maps and frequency resolution in spectrogram conversion, can significantly impact model performance. The paper identifies SOTA settings, particularly for tasks involving spectrogram perception under challenging SNR conditions.

Practical and Theoretical Implications

RFUAV provides a foundational resource for advancing UAV detection technology, with significant implications for security and privacy in smart cities and related applications. Practically, the dataset facilitates the development of more accurate and reliable RF-based detection systems. Theoretically, this work sets a precedent for integrating deep learning with complex signal processing tasks, advancing both fields.

Future Directions

The researchers acknowledge the potential for further improvements, such as increasing the number of UAV types in the dataset beyond 54 by integrating it with existing datasets like DroneRFa. The ongoing update of RFUAV with new drone data will ensure its continued relevance. Future studies might focus on evolving architectural models and signal processing techniques to enhance detection reliability.

In conclusion, RFUAV represents a significant step forward in UAV detection and identification, providing an extensive, publicly available dataset that refines current methodologies and sets the stage for future advancements in RF-based detection technologies.