A lightweight multi-attention and context fusion network for small object detection in UAV images.
Saved in:
| Title: | A lightweight multi-attention and context fusion network for small object detection in UAV images. |
|---|---|
| Authors: | Cheng, Jian1 (AUTHOR), Li, Tiansong1 (AUTHOR) tiansongli@cqnu.edu.cn, Cui, Shaoguo1 (AUTHOR), Wang, Hongkui2 (AUTHOR), Yu, Li3 (AUTHOR) |
| Source: | Displays. Jul2026, Vol. 93, pN.PAG-N.PAG. 1p. |
| Subjects: | Object recognition (Computer vision), Feature extraction, Drone photography |
| Abstract: | Small object detection in unmanned aerial vehicle (UAV) images confronts significant challenges attributed to minuscule target scales, complex backgrounds, and dense spatial distribution. Existing methods are often plagued by constraints including a limited receptive field, weak spatial awareness, and inflexible feature fusion, resulting in inadequate target representation and impaired localization accuracy. To address these issues, we propose MACF-YOLO, an efficient YOLO-based detector integrating multi-attention mechanisms and contextual fusion techniques, featuring three core modules: Multi-scale Hierarchical Perception (MHP), Coordinate Attention Enhancement (CAE), and Bi-Attention Fusion Module (BAF). The MHP module expands the perceptual range by extracting features via multi-branch dilated convolutions and facilitating hierarchical feature interaction, preserving fine-grained details while enhancing the semantic representation of small objects. The CAE module leverages complementary pooling to capture coordinate-wise dependencies, thereby boosting spatial sensitivity and suppressing background noise. The BAF module employs a collaborative strategy between spatial and channel attention to optimize multi-scale feature fusion, reinforcing salient regions across different scales. Extensive experiments conducted on the VisDrone2019 and UAVDT datasets demonstrate that MACF-YOLO consistently outperforms state-of-the-art lightweight detectors, achieving 4.4% and 6.5% gains in mAP and AP50 over YOLOv8-M on VisDrone2019, and a 1.4% higher mAP than YOLOv8-S on UAVDT, all under comparable computational budgets. The source code will be publicly available at: https://github.com/ChengJianV5/MACF-YOLO. • Lightweight MACF-YOLO for small object detection in UAV images. • Multi-scale Hierarchical Perception preserves fine details via dilated convolutions. • Coordinate Attention Enhancement improves spatial awareness with dual-path attention. • Bi-Attention Fusion enables adaptive feature fusion through spatial-channel synergy. cross-level feature fusion. • Achieves state-of-the-art results on VisDrone2019 and UAVDT with low parameters. [ABSTRACT FROM AUTHOR] |
| Copyright of Displays is the property of Elsevier B.V. and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.) | |
| Database: | Engineering Source |
Be the first to leave a comment!
