The increasing prevalence of drones in various sectors has created a pressing need for efficient and accurate detection systems to ensure airspace safety and security. This paper proposes a novel drone detection framework that combines the state-of-the-art YOLOv8 object detection model with Enhanced Super-Resolution Generative Adversarial Networks (ESRGAN) for improved visual clarity and detection accuracy. The system is capable of processing both images and videos, integrating a dynamic zoom functionality to focus on regions of interest for enhanced detection precision. By applying ESRGAN-based super-resolution enhancement on zoomed image segments, the framework effectively mitigates challenges posed by low-resolution, small object sizes, and environmental noise.
The pipeline leverages deep learning models implemented with Python, TensorFlow Hub, and the Ultralytics YOLO library, providing a practical solution for real-time drone surveillance. Experimental results demonstrate the framework’s ability to detect drones accurately under various conditions, offering promising applications in security, border control, and urban airspace management.
Introduction
Unmanned Aerial Vehicles (UAVs or drones) are increasingly used in civilian and military applications, offering benefits but also posing security and privacy risks, especially from unauthorized incursions into sensitive areas. Detecting drones across various altitudes is a critical challenge. Vision-based models like YOLO perform well at low altitudes but struggle with distant, low-resolution images.
This study proposes a multi-layered detection system combining:
YOLOv8 for low-altitude, real-time drone detection,
ESRGAN (Enhanced Super-Resolution GAN) to improve image quality and enhance detection of small or blurry drones,
Radar signal processing with deep learning for high-altitude detection,
IoT sensor fusion using acoustic and motion data to increase reliability.
The system processes input images/videos, enhances drone visibility via ESRGAN, detects drones with YOLOv8, and applies dynamic zoom to focus on regions of interest. Implemented in Python with a web interface, it supports real-time performance with acceptable latency on GPU hardware.
Key findings:
ESRGAN-enhanced images significantly improve YOLOv8 detection accuracy for small and distant drones.
Detection accuracy (mAP) increased by 12–15% with ESRGAN integration.
Improved object localization (IoU) and higher confidence scores were observed.
Slight increase in processing time remains within real-time limits.
This hybrid approach provides a scalable, robust drone detection framework suitable for surveillance and defense applications, effectively addressing low- and high-altitude detection challenges.
Conclusion
This project proposed an effective drone detection system using ESRGAN-enhanced YOLOv8 for lowaltitude vision-based detection, and radar-deep learning fusion for high-altitude tracking. Results showed improved detection accuracy and clarity, especially for small or blurry drones. The system demonstrates strong potential for real-time surveillance and security applications. Future work can focus on real-time optimization and IoT-based multi-sensor integration.
References
[1] Bochkovskiy, C. Y. Wang, and H. Y. M. Liao, YOLOv4: Optimal Speed and Accuracy of Object Detection, arXiv preprint arXiv:2004.10934, 2020.
https://arxiv.org/abs/2004.10934
[2] C. Y. Wang, A. Bochkovskiy, and H. Y. M. Liao, YOLOv7: Trainable Bag-of-Freebies Sets New State-of-the-Art for Real-Time Object Detectors, 2023.
https://github.com/WongKinYiu/yolov7
[3] C. Ledig et al., Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network, in Proc. CVPR, 2017, pp. 4681–4690.
https://arxiv.org/abs/1609.04802
[4] X. Wang et al., ESRGAN: Enhanced Super-Resolution Generative Adversarial Networks, ECCV Workshops, 2018.https://arxiv.org/abs/1809.00219
[5] Ultralytics, YOLOv8 Official Documentation, 2024.https://docs.ultralytics.com
[6] J. Redmon and A. Farhadi, YOLOv3: An Incremental Improvement, arXiv preprint arXiv:1804.02767, 2018.https://arxiv.org/abs/1804.02767
[7] Ultralytics, YOLOv8 GitHub Repository, 2024.https://github.com/ultralytics/ultralytics
[8] Google AI Blog, Understanding and Improving Object Detection, 2018.https://ai.googleblog.com/2018/07/understanding-and-improving-object.html
[9] OpenCVDev Team, OpenCV Python Documentation, 2024.
https://docs.opencv.org/
[10] K. Zhang, W. Zuo, Y. Chen, D. Meng, and L. Zhang, Beyond a Gaussian Denoiser: Residual Learning of Deep CNN for Image Denoising, IEEE Transactions on Image Processing, vol. 26, no. 7, pp. 3142–3155, 2017.
[11] T. Y. Lin, P. Goyal, R. Girshick, K. He, and P. Dollar, Focal Loss for Dense Object Detection, in Proc. ICCV, 2017, pp. 2980–2988.
[12] NVIDIA, Jetson Nano Developer Guide, 2023.
https://developer.nvidia.com/embedded/jetson-nano-developer-kit
[13] P. Oza and V. M. Patel, Robust Drone Detection Using Deep Learning Techniques: A Survey, IEEE Access, vol. 9, pp. 99332–99352, 2021.