Synthetic RF Spectrograms and IQ Signals for UAV Detection (2.4 GHz band)

This dataset provides a comprehensive collection of radio frequency (RF) signals captured in the 2.4 GHz band for Unmanned Aerial Vehicle (UAV) detection and wireless communication classification. It contains 6,000 spectrogram images (.png) and 22,375 annotated instances across three distinct classes: Drone_Signal (DJI Phantom 4 Pro), WiFi (active and idle), and Bluetooth (frequency-hopping spread spectrum). The dataset is uniquely structured to support dual analysis pipelines. It includes both the processed time-frequency representations and the raw baseband In-phase/Quadrature (IQ) signals. The IQ signals are provided as NumPy arrays (.npy) generated through synthetic IQ generation frameworks based on the source time-frequency representations. Data acquisition was performed using a USRP B210 software-defined radio with a DJI Phantom 4 Pro as the transmitter, operating at a center frequency of 2.438 GHz. The sampling rate is primarily 56 MS/s, with variations at 30 and 40 MHz, and a receiver gain of 45 dB. The spectrograms were generated using a Short-Time Fourier Transform (STFT) with NFFT=512 and an overlap of 256, utilizing the Viridis colormap. Data was collected in both indoor (semi-controlled, high SNR) and outdoor (uncontrolled, multipath, low SNR) environments. The dataset folders are organized as follows: - iq_signals/: Raw and synthetic IQ signal files in NumPy (.npy) format. - signal_crops/ and images/: Spectrogram images in (.png) format. - labels/: YOLO-format annotation files (.txt). - classes.txt: Lists class IDs (0, 1, 2). - README.txt and License.txt: Detailed metadata and licensing information. Each annotation file follows the standard YOLO object detection format (<class_id> <x_center> <y_center> <width> <height>). The images are directly compatible with YOLOv5 and YOLOv8 for object detection tasks, while the raw IQ signals are suitable for 1D Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM) networks, and signal processing experiments. This dataset serves as a valuable resource for developing robust machine learning models for RF sensing, spectrum monitoring, and anti-drone security systems.