摘要
为解决无人机目标检测中小尺寸、密集目标检测困难及在边缘设备部署困难的问题,提出了小目标检测模型Drone-YOLO。首先,提出了MF-FPN网络,在降低模型复杂度的同时融合高级语义与低级几何特征;其次,为解决小目标、密集目标难以检测问题提出了小目标检测头;而后,提出轻量化检测头LSCD,通过共享卷积降低模型复杂度,并利用组归一化提升检测性能;最后,引入Inner-WIoU损失函数,动态调整锚框权重,使模型更专注于中等质量锚框优化,从而提升回归效率与泛化能力。在公开数据集VisDrone2019上进行实验,改进后模型的mAP 0.5达到44.3%,较YOLOv11s提升6.4个百分点,参数量减少67.5%。
To address the challenges of small and dense target detection in drone applications and model deployment on edge devices,this paper presents a small-target detection model,Drone-YOLO.Firstly,a Multi-scale Feature Fusion Pyramid Network(MF-FPN)is introduced to reduce model complexity while integrating high-level semantic and low-level geometric features.Next,a small-target detection head is added to improve detection of small and dense targets.A Lightweight Shared Convolutional Detection(LSCD)head is proposed to further reduce the complexity by sharing convolutions,with group normalization enhancing performance.Lastly,the Inner-WIoU loss function is introduced to dynamically adjust anchor box weights and focus on medium-quality anchors,thus to improve regression efficiency and generalization capability.Experiments conducted on the publicly available VisDrone2019 dataset show that,the improved model achieved an mAP 0.5 of 44.3%,increased by 6.4 percentage points compared with YOLOv11s,while the number of parameters is reduced by 67.5%.
作者
牟毅
黄海松
李宜汀
付盛伟
李科
朱云伟
MU Yi;HUANG Haisong;LI Yiting;FU Shengwei;LI Ke;ZHU Yunwei(School of Mechanical Engineering,Guizhou University,Guiyang 550000,China;Key Laboratory of Advanced Manufacturing Technology,Ministry of Education,Guizhou University,Guiyang 550000,China;College of Big Data Statistics,Guizhou University of Finance and Economics,Guiyang 550000,China;Guizhou Equipment Manufacturing Digital Workshop Modeling andSimulation Engineering Research Center,Guiyang 550000,China;Information Engineering Institute,Chongqing Vocational and Technical University of Mechatronics,Chongqing 402000,China)
出处
《电光与控制》
北大核心
2026年第1期51-57,共7页
Electronics Optics & Control
基金
国家自然科学基金(52165063)
贵州省科学技术基金(黔科合平台人才-GCC〔2022〕006-1号,黔科合平台人才-CXTD〔2023〕007号)
贵州省重点科技研发计划(黔科合支撑〔2022〕165号和008号,黔科合支撑〔2023〕348号和309号,黔科合支撑〔2024〕093号)
重庆市自然科学基金(CSTB2022NSCQ-MSX1600)。
