Article Details

Abstract

This paper proposes the SOD-YOLO model, a method for precise localization and recognition of small dense objects in drone aerial images by integrating multi-scale global perception, content-aware upsampling, and cross-layer semantic enhancement. Addressing challenges such as low pixel coverage, blurred details, and severe overlapping occlusions in small objects within drone aerial images, the approach first employs Swin Transformer-Tiny as the backbone network to enhance the ability to model long-range dependencies. The CARAFE operator replaces the conventional upsampling module to optimize detail recovery and boundary preservation in feature maps. The C2f_RVB architecture enables cross-level feature fusion, significantly enhancing feature representation for multi-scale targets. Experiments results validate the model's performance on the VisDrone2019 dataset, where SOD-YOLO achieved 39.6% mAP and 44.4% F1-score, significantly outperforming mainstream models like YOLOv8 and YOLOv10. Ablation studies demonstrate that all three improvements positively contribute to detection performance. Furthermore, based on SOD-YOLO, this paper develops a dense Small Object Detection platform for urban low-altitude traffic scenarios. It supports offline analysis and real-time monitoring, featuring excellent environmental adaptability, effectively addressing complex Small Object Detection demands in urban low-altitude environments.

Keywords

low-altitude economy; drone aerial image; Small Object Detection; YOLOv8; feature fusion

References

1. [1] Wu Yiquan, Tong Kang Research progress on small object detection in unmanned aerial vehicle aerial images based on deep learning [J]. Acta Aeronautica Sinica, 2025, 46 (3): 181-207
2. [2] Populus euphratica, Zhang Tiankui, Li Bo, etc A review of research on integrated networking and technology of communication perception enabled by drones [J]. Journal of Electronics and Information Technology, 2025, 47 (4): 859-875
3. [3] Yuan Yubin, Wu Yiquan, Zhao Langyue, etc Research progress on multi object detection and tracking of unmanned aerial vehicle aerial videos based on deep learning [J]. Acta Aeronautica Sinica, 2023, 44 (18): 6-36
4. [4] Wu Han, Sun Hao, Liu Kui, etc Research progress on multi-target feature correlation technology in drone videos [J]. Acta Aeronautica Sinica, 2026, 47 (4): 131-161
5. [5] Zeng Fanshuang, Zhang Naiwen, Li Yingzhuo, etc Ground small target detection system based on YOLOv8 model and Streamlit [J]. Modern Information Technology, 2025, 9 (19): 57-61, 69
6. [6] Zeng Jing, Liao Shuzhen, Zhou Yongfu, etc Helmet detection and recognition system based on YOLOv10+PyQt5 [J]. Modern Information Technology, 2025, 9 (19): 22-25
7. [7] Cheng Jiapei, Wang Xusheng, Lin Qibin, etc Improving pedestrian and vehicle multi-target recognition, tracking and detection under YOLOv5 [J]. Modern Information Technology, 2025, 9 (17): 73-77, 82
8. [8] Li Pan, Lou Li Hybrid wafer defect pattern recognition based on Vision Transformer [J]. Modern Information Technology, 2025, 9 (19): 26-30
9. [9] Zhong Shuai, Wang Liping MCS-RETR: Improved RT-DETR based object detection method for unmanned aerial vehicle aerial images [J]. Acta Aeronautica Sinica, 2025, 46 (22): 260-276
10. [10] Tu Yuzhi, Wang Faxiang, Wu Chunlin A lightweight unmanned aerial vehicle aerial small target detection model integrating multiple attention mechanisms [J]. Computer Engineering and Applications, 2025, 61 (11): 93-104
11. [11] Zhang Xuanyu, Zhou Sihang, Huang Jian, etc Small target detection algorithm for unmanned aerial vehicle aerial photography based on high-order spatial feature extraction [J]. Computer Engineering and Applications, 2025, 61 (12): 210-221
12. [12] Chen Zhiwang, Xiao Dichuang, Lv Changhao, etc UAV aerial target detection based on multi-scale fusion and high-resolution feature enhancement [J]. Control and Decision, 2025, 40 (7): 2290-2299
13. [13] Zheng Zhanji, Feng Changkui, Zhao Yangyang, etc Non motorized vehicle small target detection method in dense scenes from the perspective of drone aerial photography [J]. Transportation Systems Engineering and Information, 2025, 25 (4): 147-161
14. [14] VARGHESE R，SAMBATH M .YOLOv8: A Novel Object Detection Algorithm with Enhanced Performance and Robustness [J].2024 International Conference on Advances in Data Engineering and Intelligent Computing Systems (ADICS)，2024：1-6.
15. [15] 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，2022,44（11）：7380-7399.
16. [16] Liu Chenjie, Liu Wei, Yang Wendi, etc DEPA-YOLO: Small Object Detection Model from the Perspective of Drones [J]. Computer Science and Exploration, 2026, 20 (1): 266-279

Funding

Science and Technology Research and Development Program of Henan Province (252102211026); National College Student Innovation and Entrepreneurship Training Program Project (202510475029)

About the authors

Wei Yuan (2004-), female, Han ethnicity, from Zhengzhou, Henan Province, undergraduate studies, research direction: object detection based on deep learning, etc; Corresponding author: Zhang Fanghui (1990-), female, Han ethnicity, from Anyang, Henan Province, lecturer, PhD, research interests: computer vision, object detection, anomaly detection, etc.