An Improved Small Object Detection Model Based on YOLOv8 for UAV Vision

doi:10.12068/j.issn.1005-3026.2025.20240116

Abstract

Abstract:

In view of easy false detection and missed detection of small objects in unmanned aerial vehicle （UAV） aerial images， as well as the requirements for real-time performance and lightweight design in UAV detection tasks， an improved lightweight and efficient model based on YOLOv8 was proposed. Firstly， the Neck part of YOLOv8 was simplified into a feature pyramid network， enabling the model to effectively utilize the detailed information extracted by shallow networks. Meanwhile， a feature fusion module was added to provide more favorable features for small object detection to the Head layer. Secondly， an efficient local attention （ELA） mechanism was integrated into the Backbone part to achieve accurate localization of target regions. Experimental results show that compared with YOLOv8s， the parameters and model size of the improved model are reduced by 50%， while the mAP_0.5 and detection speed are improved by 4%. This improved model provides a new idea for the deployment of UAV detection.

Key words: small object detection, YOLOv8, efficient local attention mechanism, lightweight model, UAV aerial photography

CLC Number:

TP 391.4

Ji-hong LIU, Rui-rui SHI. An Improved Small Object Detection Model Based on YOLOv8 for UAV Vision[J]. Journal of Northeastern University(Natural Science), 2025, 46(12): 29-37.

Figures/Tables 10

References 24

[1]	Vaswani A， Shazeer N， Parmar N， et al. Attention is all you need［J］. Advances in Neural Information Processing Systems， 2017， 30： 6000-6010.
[2]	Carion N， Massa F， Synnaeve G， et al. End-to-end object detection with Transformers［C］//Proceedings of the European Conference on Computer Vision （ECCV）. Cham： Springer， 2020： 213-229.
[3]	Girshick R， Donahue J， Darrell T， et al. Rich feature hierarchies for accurate object detection and semantic segmentation［C］//2014 IEEE Conference on Computer Vision and Pattern Recognition（CVPR）. Piscataway： IEEE， 2014： 580-587.
[4]	Ren S Q， He K M， Girshick R， et al. Faster R-CNN： towards real-time object detection with region proposal networks［J］. IEEE Transactions on Pattern Analysis and Machine Intelligence， 2016， 39（6）： 1137-1149.
[5]	Lin T Y， Dollár P， Girshick R， et al. Feature pyramid networks for object detection［C］//2017 IEEE Conference on Computer Vision and Pattern Recognition （CVPR）. Honolulu， 2017： 936-944.
[6]	Redmon J， Divvala S， Girshick R， et al. You only look once： unified， real-time object detection［C］//2016 IEEE Conference on Computer Vision and Pattern Recognition （CVPR）. Las Vegas， 2016： 779-788.
[7]	Liu W， Anguelov D， Erhan D， et al. SSD： single shot multibox detector［C］//Proceedings of the European Conference on Computer Vision （ECCV）. Cham： Springer， 2016： 21-37.
[8]	Farhadi A， Redmon J. YOLOv3： an incremental improvement［C］//2018 IEEE Conference on Computer Vision and Pattern Recognition（CVPR）. Salt Lake City， 2018： 1-6.
[9]	Glenn J. YOLOv5 release v7.0 ［EB/OL］（2022-11-22）［2024-03-10］. .
[10]	Law H， Deng J. CornerNet： detecting objects as paired keypoints［C］//Proceedings of the European Conference on Computer Vision （ECCV）. Cham： Springer， 2018： 765-781.
[11]	Tian Z， Shen C H， Chen H， et al. FCOS： fully convolutional one-stage object detection［C］// International Conference on Computer Vision （ICCV）. Seoul， 2019： 9626-9635.
[12]	Glenn J. YOLOv8 release v8.1.0.［EB/OL］. （2024-01-10）［2024-03-10］. .
[13]	Wang C Y， Yeh I H， Liao H Y M. YOLOv9： learning what you want to learn using programmable gradient information［C］//Proceeding of the European Conference on Computer Vision（ECCV）. Cham： Springer， 2024： 1-21.
[14]	Li C L， Yang T， Zhu S J， et al. Density map guided object detection in aerial images［C］//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops （CVPRW）. Seattle， 2020： 737-746.
[15]	燕舒乐，陈润宇，蔡念，等.基于多跳深度网络的红外微小目标检测方法［J］.激光与光电子学进展， 2024， 61（22）： 2237008.
	Yan Shu-le， Chen Run-yu， Cai Nian， et al. Infrared small target detection method based on multi-hop depth network［J］. Laser & Optoelectronics Progress， 2024， 61（22）： 2237008.
[16]	吴海斌，张亚，胡鹏.面向无人机航拍图像小目标检测方法［J］.安徽工业大学学报（自然科学版），2024，41（1）：65-73.
	Wu Hai-bin， Zhang Ya， Hu Peng. A small target detection method for unmanned aerial vehicle aerial photography images［J］. Journal of Anhui University of Technology （Natural Science）， 2024， 41（1）： 65-73.
[17]	Meethal A， Granger E， Pedersoli M. Cascaded zoom-in detector for high resolution aerial images［C］//2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops （CVPRW）. Vancouver， 2023： 2046-2055.
[18]	Liu S， Qi L， Qin H F， et al. Path aggregation network for instance segmentation［C］//2018 IEEE/ Conference on Computer Vision and Pattern Recognition（CVPR）. Salt Lake City， 2018： 8759-8768.
[19]	Xu W， Wan Y. ELA： efficient local attention for deep convolutional neural networks ［EB/OL］. （2024-03-02）［2024-04-10］. .
[20]	Tan M X， Pang R M， Le Q V. EfficientDet： scalable and efficient object detection［C］//2020 IEEE Conference on Computer Vision and Pattern Recognition （CVPR）. Seattle， 2020： 10778-10787.
[21]	Hou Q B， Zhou D Q， Feng J S. Coordinate attention for efficient mobile network design［C］//2021 IEEE Conference on Computer Vision and Pattern Recognition （CVPR）. Nashville，2021： 13708-13717.
[22]	Hou Q B， Zhang L， Cheng M M， et al. Strip pooling： rethinking spatial pooling for scene parsing［C］//2020 IEEE Conference on Computer Vision and Pattern Recognition （CVPR）. Seattle， 2020： 4002-4011.
[23]	Zhu P F， Wen L Y， Du D W， et al. Detection and tracking meet drones challenge［J］. IEEE Transactions on Pattern Analysis and Machine Intelligence， 2021， 44（11）： 7380-7399.
[24]	Zou Z X， Chen K Y， Shi Z W， et al. Object detection in 20 years： a survey［J］. Proceedings of the IEEE， 2023， 111（3）： 257-276.

环境	型号
GPU	NVIDIA GeForce RTX 4090
内存	24 GB
操作系统	Windows
系统位数	64
编译器	Pycharm 2023
CUDA	CUDA11.8
编译环境	Python3.9.19
深度学习框架	PyTorch2.2.2

环境	型号
GPU	NVIDIA GeForce RTX 4090
内存	24 GB
操作系统	Windows
系统位数	64
编译器	Pycharm 2023
CUDA	CUDA11.8
编译环境	Python3.9.19
深度学习框架	PyTorch2.2.2

YOLOv8版本	深度倍数	宽度倍数	比率
YOLOv8n	0.33	0.25	2.0
YOLOv8s	0.33	0.50	2.0
YOLOv8m	0.67	0.75	1.5
YOLOv8l	1.00	1.00	1.0
YOLOv8x	1.00	1.25	1.0

YOLOv8版本	深度倍数	宽度倍数	比率
YOLOv8n	0.33	0.25	2.0
YOLOv8s	0.33	0.50	2.0
YOLOv8m	0.67	0.75	1.5
YOLOv8l	1.00	1.00	1.0
YOLOv8x	1.00	1.25	1.0

参数	数值
训练轮数	250
批次大小	8
图像尺寸/像素	640×640
交并比	0.7
初始学习率	1×10^-2
最终学习率	1×10^-4
数据增强	1.0
动量	0.937