TB-YOLOv8:An Algorithm of Detecting Object Turned in the Direction of Drone Aerial Imagery
XIA Zihang
YUE Yumei
JI Shude
REN Zhaoxu
ZHANG Ruizhe
Abstract：Aimed at the problems that object is dense in distribution,small object is easy to omit,and scale is variable in drone aerial imagery,an improved algorithm TB-YOLOv8 is proposed.This method is to incorporate a triplet attention-based C2f_T module into the backbone to enhance shallow feature extrac-tion and key region awareness,reducing false positives and missed detections.A lightweight BiFPN is in-tegrated into the neck to improve multi-scale feature fusion efficiency.And simultaneously,the WIoUv3 loss function is introduced to enhance model robustness and localization accuracy.The experimental results on the VisDrone2019 dataset show that the TB-YOLOv8 achieves a 3.0％ improvement in mAP,a 6.31％reduction in parameter count,and a detection speed of 164.1 FPS compared to the original YOLOv8s mod-el.Moreover,the TB-YOLOv8 is prior to the Faster R-CNN,the RT-DETR,the YOLOv7,and the YOLOv11 in terms of detection performance,has high accuracy,efficiency,and real-time capability,and there is considerable potential in aerial target detection application.
Keywords：object detectiondrone aerial imageryYOLOv8
Publication Date：2025-12-25
Online Publishing Date：2025-12-26(First online date of this platform, not the publication date of the document)
Pages：10( 96-105 )