WDFS-DETR: A Transformer-based framework with multi-scale attention for small object detection in UAV Engineering Tasks

计算机科学 变压器 比例(比率) 系统工程 人工智能 工程类 地图学 地理 电气工程 电压
作者
Jinjiang Liu,Yonghua Xie
出处
期刊:Results in engineering [Elsevier BV]
卷期号:27: 105930-105930 被引量:6
标识
DOI:10.1016/j.rineng.2025.105930
摘要

• WDFS-DETR improves small object detection in UAV imagery via wavelet-based attention. • DSAN dynamically switches normalization modes to boost both training and inference efficiency. • FFDPN aligns multi-scale features for better robustness in cluttered aerial scenes. • Achieves 50.1 FPS on Jetson Orin Nano while surpassing RT-DETR and YOLOv8l accuracy. • Demonstrates strong cross-dataset generalization on VisDrone2019, UAVDT, and DOTA. With the growing adoption of Unmanned Aerial Vehicles (UAVs) in surveillance, emergency response, and environmental monitoring, accurate small object detection under resource constraints remains a critical challenge. To address this issue, we propose WDFS-DETR (Wavelet-based Dual-stage Feature-enhanced Detection Transformer), a Transformer-based detection framework built upon RT-DETR (Real-Time Detection Transformer). The framework integrates four custom-designed components to jointly enhance detection accuracy and efficiency. First, BasicBlock-WTCM (Wavelet-based Transform Coordinate Module) improves small object perception by modeling spatial and channel semantics across scales. Second, the Dual-Stage Adaptive Normalization (DSAN) dynamically selects appropriate normalization strategies during training and inference to improve convergence and runtime performance. Third, the Feature-Focused Diffusion Pyramid Network (FFDPN) enhances context modeling and robustness via hierarchical multi-scale feature alignment and fusion. Finally, the boundary-aware Slide-VarifocalLoss combines sliding window mechanisms with class-weighted reweighting to address class imbalance and improve boundary localization. Experiments show that WDFS-DETR improves mAP@0.5 by 2.2 % over RT-DETR-r18 on VisDrone2019. When deployed on Jetson Orin Nano, inference speed improves from 38.6 FPS to 50.1 FPS, highlighting its suitability for real-time deployment on lightweight platforms. The model also generalizes well to UAVDT and DOTA datasets, demonstrating its applicability to real-world UAV-based detection tasks in complex engineering settings. Source code is available at: https://github.com/liuliuliu2002/WDFS-DETR .
最长约 10秒,即可获得该文献文件

科研通智能强力驱动
Strongly Powered by AbleSci AI
科研通是完全免费的文献互助平台,具备全网最快的应助速度,最高的求助完成率。 对每一个文献求助,科研通都将尽心尽力,给求助人一个满意的交代。
实时播报
刚刚
五四三二一完成签到,获得积分10
刚刚
刚刚
徐霖发布了新的文献求助10
1秒前
怪僻完成签到,获得积分10
1秒前
1秒前
袋袋完成签到,获得积分10
1秒前
1秒前
跳跃以山的应助被fly采纳,获得10
1秒前
钌铑钯完成签到,获得积分10
1秒前
1秒前
最哈的蜜瓜完成签到,获得积分10
1秒前
FYhan发布了新的文献求助10
1秒前
2秒前
青山发布了新的文献求助10
2秒前
2秒前
勤恳大树发布了新的文献求助15
2秒前
风格举报韩小小的求助涉嫌违规
3秒前
小张完成签到 ,获得积分10
3秒前
大橘发布了新的文献求助10
3秒前
4秒前
xjt发布了新的文献求助10
4秒前
liangmh完成签到,获得积分10
4秒前
孤独的迎滑完成签到,获得积分10
4秒前
Sano完成签到,获得积分10
4秒前
小小科研人完成签到,获得积分10
4秒前
4秒前
Hello的应助被蓓蓓采纳,获得10
5秒前
1001完成签到,获得积分10
5秒前
r...........发布了新的文献求助10
5秒前
小蛤蟆完成签到,获得积分10
5秒前
秒速五厘米的应助被modoun采纳,获得10
6秒前
aaronzhu1995完成签到,获得积分10
6秒前
冷傲书蝶完成签到,获得积分10
6秒前
sink发布了新的文献求助10
6秒前
kk完成签到,获得积分10
6秒前
想回家发布了新的文献求助10
6秒前
哈哈哈完成签到,获得积分10
6秒前
ada发布了新的文献求助10
6秒前
7秒前
高分求助中
(应助此贴封号)【重要!!请各用户(尤其是新用户)详细阅读】【科研通的精品贴汇总】 10000
拉伸高分子熔体中缠结行为的模拟和小角中子散射研究 1500
The World Atlas of Wine 9th edition 1000
奈特基础组织学:配套组织病理学(第4版)(Netter's Essential Histology With Correlated Histopathology) 800
John McDowells Bildungstheorie 800
八旗制度の研究 600
Analyses de reconnaissances de dettes passées devant les échevins d'Ypres (1249-1291) 600
热门求助领域 (近24小时)
化学 材料科学 医学 生物 纳米技术 计算机科学 工程类 有机化学 化学工程 生物化学 物理 复合材料 内科学 催化作用 光电子学 人工智能 细胞生物学 无机化学 物理化学 数学
热门帖子
关注 科研通微信公众号,转发送积分 8159581
求助须知:如何正确求助?哪些是违规求助? 9555100
关于积分的说明 21724719
捐赠科研通 7749543
什么是DOI,文献DOI怎么找? 3427281
关于科研通互助平台的介绍 2552278
邀请新用户注册赠送积分活动 2471619