全文下载:
202604017.pdf
文章编号:1672-6987(2026)04-0136-09;DOI:10. 16351/j. 1672-6987. 2026. 04. 017
杨继强,高 林* (青岛科技大学 自动化与电子工程学院,山东 青岛 266061)
摘 要:无人机航拍图像中小目标特征信息少、数量多、易受环境影响导致小目标检测效果不 佳。针对该问题,提出了一种无人机视角下的小目标检测算法 STD-YOLO。首先,为增强对 目标的表征和定位,研究并设计了 PCAM 注意力模块嵌入到颈部网络中,以提高小目标特征 在通道和空间维度上的表达能力;其次,为减少在特征提取网络中目标信息的丢失构建了多 扩展卷积模块 MCM,提高目标感受野,丰富上下文信息;最后,在特征融合网络中设计了多尺 度特征融合结构,融合低层目标信息和高层语义信息,并使用 4 个不同尺度的检测头以增强对 小目标的感知能力;实验结果表明,STD-YOLO 对小目标的检测能力有明显提升,mAP0. 5 较 原 YOLOv5 模型提升了 6. 8%,性能优于目前主流模型,能够有效应用在无人机航拍图像中的 小目标检测任务。
关键词:机器视觉;无人机航拍;注意力模块;多扩展卷积模块;小目标检测头
中图分类号:TP 391. 4 文献标志码:A
引用格式:杨继强,高林 . 面向无人机航拍的小目标检测算法[J]. 青岛科技大学学报(自然 科学版),2026,47(4):136-144.
YANG Jiqiang, GAO Lin. Small target detection algorithm for UAV aerial photography[J]. Journal of Qingdao University of Science and Technology(Natural Science Edition),2026,47 (4):136-144.
Small Target Detection Algorithm for UAV Aerial Photography
YANG Jiqiang,GAO Lin (College of Automation and Electronic Engineering,Qingdao University of Science and Technology,Qingdao 266061, China)
Abstract:Due to the lack of feature information, large quantity, and susceptibility to environ‑ mental influences of small objects in UAV aerial images, the detection effect of small objects is poor. To address this problem, an STD-YOLO small object detection algorithm under UAV perspective is proposed. Firstly, in order to enhance the representation and localization of objects, the PCAM attention module is researched and designed to be embedded into the neck network to improve the representation ability of small object features in channel and spatial dimensions. Secondly, to reduce the loss of target information in the feature extraction net‑ work, the multi-expansion convolution module MCM is constructed to increase the receptive field of the target and enrich contextual information. Finally, a multi-scale feature fusion struc‑ ture is designed in the feature fusion network to fuse low-layer target information and highlayer semantic information, and 4 detection heads of different scales are used to enhance the perception ability of small objects. Experiments show that STD-YOLO has significantly improved detection capability for small objects, increased by 6. 8% compared to the original YOLOv5 model, outperforming mainstream models, and can effectively apply to small object detection in UAV aerial images.
Key words:machine vision; drone aerial photography; attention module; multi-expansion con‑ volution module; small target detection head
收稿日期:2025-06-10
基金项目:山东省自然科学基金项目(ZR2021MF023).
作者简介:杨继强(1998—),男,硕士研究生 . * 通信联系人 .