Hybrid-YOLO: A Lightweight UAV Small Object Detection Network based on Traditional Visual Feature Compensation and Lightweight Architecture

Haoyu Zhang

doi:10.47297/taposatWSP2633-456910.20260704

PDF
导出
分享
收藏
专辑

Hybrid-YOLO: A Lightweight UAV Small Object Detection Network based on Traditional Visual Feature Compensation and Lightweight Architecture

7卷第4期

浏览量: 0 Downloads: 0

Hybrid-YOLO: A Lightweight UAV Small Object Detection Network based on Traditional Visual Feature Compensation and Lightweight Architecture
Hybrid-YOLO: A Lightweight UAV Small Object Detection Network based on Traditional Visual Feature Compensation and Lightweight Architecture
Institute of Aeronautics and Astronautics, University of Electronic Science and Technology of China, Chengdu, Sichuan, 611731, China
2026年7卷第4期页码：59-66
纸质出版：2026-04-30
DOI：10.47297/taposatWSP2633-456910.20260704
Accepted：

Scan QR Code

Haoyu Zhang. (2026). Hybrid-YOLO: A Lightweight UAV Small Object Detection Network based on Traditional Visual Feature Compensation and Lightweight Architecture. 科技理论与实践（英文版）, 7(4), 59-66.

Zhang Haoyu.Hybrid-YOLO: A Lightweight UAV Small Object Detection Network based on Traditional Visual Feature Compensation and Lightweight Architecture[J].Theory and Practice of Science and Technology,2026,07(04):59-66.
Haoyu Zhang. (2026). Hybrid-YOLO: A Lightweight UAV Small Object Detection Network based on Traditional Visual Feature Compensation and Lightweight Architecture. 科技理论与实践（英文版）, 7(4), 59-66. DOI： 10.47297/taposatWSP2633-456910.20260704.

Zhang Haoyu.Hybrid-YOLO: A Lightweight UAV Small Object Detection Network based on Traditional Visual Feature Compensation and Lightweight Architecture[J].Theory and Practice of Science and Technology,2026,07(04):59-66. DOI： 10.47297/taposatWSP2633-456910.20260704.

摘要

Abstract

Unmanned Aerial Vehicles (UAVs) are increasingly utilized in low-altitude optical remote sensing

but constrained by flight altitude and broad field of view

detecting objects in UAV-captured imagery presents significant challenges due to extremely small target sizes and complex backgrounds. Existing lightweight detection networks often suffer from a severe loss of high-frequency detail features of small objects during the deep feature abstraction process caused by excessive downsampling. To address this critical bottleneck

this paper proposes a lightweight UAV small object detection network named Hybrid-YOLO. First

prior analysis of the dataset is conducted through K-Means clustering

revealing the severe scale distortion of tiny objects. Second

to compensate for feature degradation

a Visual Feature Fusion (VFF) module is designed to losslessly inject physical contours extracted by the Soft-Canny operator and Laplacian texture features into the deep semantic space at the extremely shallow layer of the backbone network. Subsequently

to overcome the extreme sensitivity of bounding box regression to geometric displacements for tiny objects

a scale-adaptive Normalized Wasserstein Distance (NWD) loss based on object pixel area is introduced. At the underlying tensor level

dimensional expansion is employed to resolve the broadcasting dimension explosion problem. Finally

a P2 tiny object detection head is specifically constructed to preserve ultra-high-resolution features

synergistically equipped with Focal Loss to suppress massive background negative samples. Experimental results on the VisDrone2019-DET dataset demonstrate that

with a highly constrained parameter size of only 2.93M

Hybrid-YOLO achieves an overall mAP@0.5 of 38.2%. Concurrently

in full-precision tests on the RTX 3090 platform

the model sustains an ultra-high inference speed of 140.46 FPS with a computational complexity of 12.5 GFLOPs

achieving a Pareto optimal balance between detection accuracy and edge deployment real-time performance.

关键词

Keywords

references

Zhu, P., Wen, L., Du, D., Bian, X., Fan, H., Hu, Q., Ling, H. (2021) Detection and tracking meet drones challenge. IEEE Transactions on Pattern Analysis and Machine Intelligence , 44(11): 7380–7399.

M. Tan, R. Pang and Q. V. Le, "EfficientDet: Scalable and Efficient Object Detection," 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , Seattle, WA, USA, 2020, pp. 10778 - 10787 .

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 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , Vancouver, BC, Canada, 2023, pp. 7464 - 7475 .

Y. Cao et al ., "VisDrone-DET2021: The Vision Meets Drone Object detection Challenge Results," 2021 IEEE/CVF International Conference on Computer Vision Workshops (ICCVW) , Montreal, BC, Canada, 2021, pp. 2847 - 2854 .

Di X, Cui K, Wang R-F. Toward Efficient UAV-Based Small Object Detection: A Ligh tweight Network with Enhanced Feature Fusion. Remote Sensing . 2025; 17(13):2235.

X. Zhu, S. Lyu, X. Wang and Q. Zhao, "TPH-YOLOv5: Improved YOLOv5 Based on Transformer Prediction Head for Object Detection on Drone-captured Scenarios," 2021 IEEE/CVF International Conference on Computer Vision Workshops (ICCVW) , Montreal, BC, Canada, 2021, pp. 2778 - 2788 .

Sunkara, R., Luo, T. (2023). No More Strided Convolutions or Pooling: A New CNN Building Block for Low-Resolution Images and Small Objects. In: Amini, MR., Canu, S., Fischer, A., Guns, T., Kralj Novak, P., Tsoumakas, G. (eds) Machine Learning and Knowledge Discovery in Databases. ECML PKDD 2022. Lecture Notes in Computer Science(), vol 13715. Springer, Cham.

Zheng Z, Zhao J, Fan J (2025) YOLO-GML: An object edge enhancement detection model for UAV aerial images in complex environments.PLOS ONE 20(7): e0328070.

Zhang Z, Yi H H, Zheng J. Focusing on small objects detector in aerial images[J]. Acta Electronica Sinica, 2023, 51(4): 944-955.

Xu, C., Wang, J., Yang, W., Yu, H., Yu, L., Xia, G.S. (2022) Detecting tiny objects in aerial images: A normalized Wasserstein distance and a new benchmark. ISPRS Journal of Photogrammetry and Remote Sensing , 190: 79–93.

Yang, X., Yan, J ., Ming, Q., Wang, W., Zhang, X., Tian, Q., 2021. Rethinking rotated object detection with gaussian wasserstein distance loss. In: Proceedings of the 38th International Conference on Machine Learning (ICML) . PMLR 139:11830-11841.

Souza, B.J., Stefenon, S.F., Singh, G., Freire, R.Z. (2023) Hybrid-YOLO for classification of insulators defects in transmission lines based on UAV. International Journal of Electrical Power & Energy Systems , 148: 108982.

Shakya, B., El-Gayar, O., Kaabi, J. (2025) Small-Object detection at the edge: A Pareto-efficient benchmark of lightweight YOLO models on UAV and overhead datasets. IEEE Access , 14: 528–548.

Liang, S., Feng, X., Xie, M., Tang, Q., Zhu, H., Li, G. (2025) Lightweight YOLO-SR: A method for small object detection in UAV aerial images. Applied Sciences , 15: 13063.

Wang A, Chen H, Liu L, et al. Yolov10: Real-time end-to-end object detection[J]. Advances in neural information processing systems, 2024, 37: 107984-108011.

Z. Li, H. Chen, Z. Wang, Y. Wang, J. Li and H. Zhu, "V10Drone: Improved YOLOv10 with KAN for Small Object Detection in Aerial Images," 2024 8th International Symposium on Computer Science and Intelligent Control (ISCSIC) , Zhengzhou, China, 2024, pp. 148 - 154 .

文章被引用时，请邮件提醒。

Submit

Research on Key Technologies of Hydrogen Fuel Cell UAV Power Systems

Survey of High-Speed Vehicle Tracking Technology for Unmanned Aerial Vehicles Based on Neural Networks and YOLO

Exploration of the Real-Time Detection Model Transformer (UAV-DETR) for Small Object Detection in UAV Imagery

A Lightweight Real-time Detection System for Power Line Defects Based on DartNet-YOLO

Construction and Application of Coastline Erosion Risk Prevention and Control System Based on Ecological Engineering

Related Author

Zhipeng Xu School of Mechanical Engineering, Hefei University of Technology, Hefei, Aanhui

Kaizhuo Liu School of Electromechanical Engineering, Guangdong University of Technology

Yican Yang School of Information Science and Engineering,Computer Science and Technology Major,Ocean University of China

Zitong Xu Jinan Xinhang Experimental Foreign Language School

Related Institution

School of Mechanical Engineering, Hefei University of Technology, Hefei, Aanhui

School of Electromechanical Engineering, Guangdong University of Technology

School of Information Science and Engineering,Computer Science and Technology Major,Ocean University of China

Jinan Xinhang Experimental Foreign Language School

⁰