Dataset_Drones_Bluetooth_Wifi
Description
This dataset contains labeled RF spectrograms for detection and classification of unmanned aerial vehicles (UAVs) operating in the 2.4 GHz ISM band under realistic WiFi and Bluetooth interference. Data were acquired using a USRP B210 software-defined radio with a center frequency of 2.438 GHz, sampling rate of 56 MS/s, and receiver gain of 45 dB. Raw IQ samples were transformed into time-frequency spectrograms using Short-Time Fourier Transform (STFT) with NFFT=512 and overlap=256. The Viridis colormap was applied. The dataset includes 6,842 spectrogram images (.png) and corresponding YOLO-format annotation files (.txt). Three signal classes are annotated: - Class 0: Drone_Signal (DJI Phantom 4 Pro, active video transmission) - Class 1: WiFi (2.4 GHz, active and idle) - Class 2: Bluetooth (frequency-hopping spread spectrum) Total annotated instances: 22,375. File structure: - images/ : all .png spectrograms - labels/ : all .txt annotation files (YOLO format: class_id x_center y_center width height) - classes.txt : lists class names (0,1,2) - README.txt : detailed description Acquisition scenarios include indoor (clean, high SNR) and outdoor (cluttered, low SNR, multipath) environments. The data are directly usable for training object detection models such as YOLOv5, YOLOv8, and other deep learning frameworks for RF-based drone detection. For full experimental details, please refer to the accompanying Data in Brief article.
Files
Steps to reproduce
1. Download the dataset and extract the .zip file. 2. The folder structure contains images/ and labels/ subfolders. 3. Use any YOLO-compatible framework to load image-annotation pairs. 4. Example training command for YOLOv8: yolo train data=dataset.yaml model=yolov8n.pt 5. A sample dataset.yaml file is provided in the root folder.