现代电子技术2025，Vol.48Issue(19)：103-109,7.DOI:10.16652/j.issn.1004-373x.2025.19.017
面向无人机航拍图像的端到端检测算法
End-to-end detection algorithm for UAV aerial photography images
摘要
Abstract
In view of the detection difficulties of UAV aerial photography images,for instance,the objects are not so big,the occlusions are serious and the backgrounds are complex,an end-to-end UAV aerial photography image detection algorithm based on improved YOLO-DETR is proposed.Firstly,the detection head of YOLOv8 is replaced with the decoder of RT-DETR to get rid of the adverse effect of NMS on small object detection.Secondly,the designed CAA-Fusion module is introduced into the backbone network,and two auxiliary information flows enhanced by CAA are merged with P4 feature map,which enriches fine-grained features and enhances the feature extraction ability of the backbone.And then,a small object detection layer is introduced into Neck,an H-GFPN structure is adopted,and cross-scale connections are introduced to enhance the information fusion among different scales,so as to improve the detection ability for small objects.Finally,C2f-FR-EMA module is introduced into Neck to improve the attention to small objects in feature fusion.In comparison with YOLOv8s model,the mAP@0.5 of the proposed algorithm is improved by 11.3%on the UAV aerial dataset VisDrone2019,and its accuracy is higher than YOLO series model and RT-DETR model which have roughly equivalent quantity of model parameters,which verifies the effectiveness of the improved algorithm in UAV aerial photography images.关键词
无人机航拍/小目标/YOLOv8/RT-DETR/辅助信息流/H-GFPN/C2f-FR-EMAKey words
UAV aerial photography/small object/YOLOv8/RT-DETR/auxiliary information flow/H-GFPN/C2f-FR-EMA分类
信息技术与安全科学引用本文复制引用
童浩然,金涵..面向无人机航拍图像的端到端检测算法[J].现代电子技术,2025,48(19):103-109,7.基金项目
国家自然科学基金项目(62227815) （62227815）
