软件导刊2026，Vol.25Issue(8)：91-98,8.DOI:10.11907/rjdk.251533
基于旋涡流卷积的轻量化航拍小目标检测算法
Lightweight Aerial Small Object Detection Algorithm Based on Vortex Flow Convolution
摘要
Abstract
The detection of small targets in aerial images poses severe challenges to the feature extraction capability of convolutional neural networks due to the small scale of the targets,complex backgrounds,and significant intra-class variations.Such intra-class variations are mainly manifested as the appearance changes of targets belonging to the same class under different imaging conditions,making feature learn-ing more challenging.Traditional object detection algorithms often suffer from reduced localization accuracy when handling aerial small ob-jects,primarily due to insufficient receptive fields or interference from background noise.This paper proposes a novel Vortex Flow Convolution(VFConv)module based on the YOLOv11 network.VFConv effectively enhances the feature saliency of small objects and suppresses back-ground interference through mechanisms such as vortex core extraction,peripheral flow suppression,and dynamic flow modulation.Integrated into the VortexBottleneck module and combined with flow density-tuned residual connections,VFConv further improves the stability and ro-bustness of feature representation.Experimental results demonstrate that YOLO11-VFConv achieves an mAP50 of 74.40％,outperforming YO-LOv11n by 7.7％,YOLOv8n by 6.5％,YOLOv10n by 10.6％,and YOLOv12n by 6.3％.With superior performance in localization accuracy,overall detection capability,and inference efficiency,YOLO11-VFConv provides an efficient and practical solution for aerial small object de-tection.关键词
卷积神经网络/小目标检测/YOLOv11/旋涡流卷积Key words
convolutional neural network/small object detection/YOLOv11/vortex flow convolution分类
信息技术与安全科学引用本文复制引用
田乔梅,刘杨,刘春红..基于旋涡流卷积的轻量化航拍小目标检测算法[J].软件导刊,2026,25(8):91-98,8.基金项目
网络与交换技术全国重点实验室开放课题(SKLNST-2024-1-11) （SKLNST-2024-1-11）
河南省科技攻关项目(242102321148) （242102321148）
