Scholarlink

<p>Malicious drones pose significant threats to sensitive sites such as airports, military bases, and nuclear power plants. Accurate detection and classification are required before any neutralization efforts. We address this challenge as a real-time object detection problem, focusing on the radio frequency (RF) signature of the communication signals exchanged between the drone and its remote controller, represented in spectrogram images. We propose a modified version of the You Only Look Once (YOLO) object detection model, RF-YOLO, and evaluate it using a customized dataset of spectrogram images generated from drone remote controller RF signals under various signal-to-noise ratio (SNR) values. Experimental results show that RF-YOLO achieves a mAP of 0.9213, precision of 0.9800, and recall of 0.9750, outperforming YOLOv3, YOLOv5, YOLOv8, RT-DETR, and RetinaNet. RF-YOLO’s mAP is 7.8%, 1.6%, and 4.7% higher than YOLOv8, RT-DETR, and RetinaNet, respectively, while also achieving superior precision and recall, demonstrating its effectiveness in minimizing false positives and maximizing true detections.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

RF-YOLO: a modified YOLO model for UAV detection and classification using RF spectrogram images

  • 期刊论文

  • 发表日期：

    2025年2月15日

查看全文
  • Tijeni Delleji,
  • Feten Slimeni

摘要

Malicious drones pose significant threats to sensitive sites such as airports, military bases, and nuclear power plants. Accurate detection and classification are required before any neutralization efforts. We address this challenge as a real-time object detection problem, focusing on the radio frequency (RF) signature of the communication signals exchanged between the drone and its remote controller, represented in spectrogram images. We propose a modified version of the You Only Look Once (YOLO) object detection model, RF-YOLO, and evaluate it using a customized dataset of spectrogram images generated from drone remote controller RF signals under various signal-to-noise ratio (SNR) values. Experimental results show that RF-YOLO achieves a mAP of 0.9213, precision of 0.9800, and recall of 0.9750, outperforming YOLOv3, YOLOv5, YOLOv8, RT-DETR, and RetinaNet. RF-YOLO’s mAP is 7.8%, 1.6%, and 4.7% higher than YOLOv8, RT-DETR, and RetinaNet, respectively, while also achieving superior precision and recall, demonstrating its effectiveness in minimizing false positives and maximizing true detections.