Hierarchical and local detail enhancement for small object detection in aerial images
Accurate small object detection in aerial images is important for vision-based applications. Detecting unmanned aerial vehicles (UAVs) and birds in airport clearance zones is particularly important for aviation safety. Nevertheless, existing approaches still suffer from pronounced semantic inconsistencies between high-level semantic representations and low-level structural cues during multi-scale feature fusion. Furthermore, local structural information of small targets is prone to degradation during deep semantic modeling and cross-scale feature aggregation, leading to compromised detection performance. To mitigate these issues, we introduce HLD-DETR, a Hierarchical and Local Detail Enhancement framework for small object detection in aerial images. In particular, the Hierarchical Detail Complementation Module (HDCM) selectively integrates low-level structural details into deep semantic features via a cross-level cross-attention mechanism. Thereafter, the Local Detail Refinement Module (LDRM) combines multidirectional local structural modeling with channel attention to explicitly refine the local structural features of small objects. Collectively, HDCM and LDRM systematically address cross-level interaction and local structural enhancement, synergistically improving small object feature representation. Additionally, we construct the UB-SOD dataset, specifically for detecting small UAVs and birds in airport clearance zones. Extensive experiments on both UB-SOD and the publicly available VisDrone2019 datasets demonstrate the effectiveness of HLD-DETR for small object detection in aerial images, while further validating its applicability to airport clearance zone scenarios.
Authors
- Shanliang Liu (ORCID: https://orcid.org/0000-0003-4723-0965)
- Guangshuai Gao (ORCID: https://orcid.org/0000-0002-5050-2311)
- Yiran Bie
- Xiaoyang Jia
- Jiangang Wang
- Baorui Wang
Institutions
- Zhongyuan University of Technology (CN)
- China Electronics Technology Group Corporation (CN)
Publication Details
- Journal
- Scientific Reports
- Published
- 2026-11-18
- DOI
- https://doi.org/10.1038/s41598-026-80751-3
- Primary Topic
- Advanced Neural Network Applications
- Type
- article
- Field-Weighted Citation Impact
- 0.00
Funders
- National Natural Science Foundation of China
- Zhongyuan University of Technology