现代电子技术2025，Vol.48Issue(23)：39-47,9.DOI:10.16652/j.issn.1004-373x.2025.23.006
FDD-YOLO多尺度特征融合航拍目标检测算法
Aerial object detection algorithm FDD-YOLO with multi-scale feature fusion
摘要
Abstract
An improved object detection algorithm FDD-YOLO based on YOLOv8n is proposed in order to eliminate the missed detections caused by multi-scale and occlusion in the task of small object detection in UAV aerial photography.A receptive-field concentration attention convolution(RFCAConv)module is introduced into the backbone network to extract deep-level features,which significantly enhances the effectiveness and accuracy of feature extraction.A cross-scale weighted feature fusion module FBGM is designed in the part of neck.This module overcomes the deficiency that small-object features are prone to being lost due to the significant disparity in object scales,which significantly improves the ability to accurately detect small objects in complex scenarios while streamlining the network structure.Finally,a task-aligned dynamic detection head(TADDH)is introduced to increase effective interaction among tasks,reducing the occurrence of false detections and missed detections.The performance of FDD-YOLO on the VisDrone2019 dataset is verified with experiments.In comparison with the original baseline model based on YOLOv8n,the precision ratio of FDD-YOLO improves by 4.9%,its recall rate by 4.0%,its mAP@0.5 by 4.9%,and its mAP@0.5:0.95 by 3.3%while its number of parameters rises by 2.7%,achieving better effects.To sum up,the proposed algorithm has more advantages in the field of UAV aerial object detection.关键词
航拍小目标检测/YOLOv8n/特征提取/特征融合/VisDrone2019/TADDHKey words
aerial small object detection/YOLOv8n/feature extraction/feature fusion/VisDrone2019/TADDH分类
电子信息工程引用本文复制引用
王松,赵辉,曲文龙,张文涛..FDD-YOLO多尺度特征融合航拍目标检测算法[J].现代电子技术,2025,48(23):39-47,9.基金项目
陕西省自然科学基础研究计划项目(2024TC-YBQN-0725) （2024TC-YBQN-0725）
