UAV small target detection algorithm integrating dynamic sparsity
XU Xianglei
XU Qinjun
HU Yongqiang
Abstract：Aiming at the problems of missed detection and false detection caused by the uneven distribution of small targets and indistinct feature information,a lightweight UAV small target detection algorithm named BEW-YOLO is proposed.The original loss function is replaced with the GWD(Gaussian Wasserstein Distance)Loss function based on Gaussian Wasserstein distance to optimize the problem of discontinuous target boundaries.A 160×160 detection head for small target detection is embedded to enhance the ability of detecting small targets.The efficient channel attention(ECA)module and dynamic sparse attention mechanism(BiFormer)are introduced to strengthen the model's ability to extract feature information of targets and improve detection accuracy.Experimental results on the VisDrone2019 dataset show that compared with the baseline algorithm YOLOv8n,the mAP@0.5 and mAP@0.5:0.9 are increased by 6.9%and 4.3%respectively,while the Precision and Recall are simultaneously improved by 5.8%and 7.3%.Generalization experiments are conducted on the RSOD and NWPU VHR-10 datasets,and the mAP@0.5(%)is increased by 2.1%and 3.5%respectively.Moreover,the total number of parameters of the model is only 3.2M,which indicates that the proposed BEW-YOLO algorithm can effectively accomplish the task of small object detection.
Keywords：small target detectionYOLOattention mechanismloss functionlightweight
Publication Date：2025-12-15
Online Publishing Date：2026-04-01(First online date of this platform, not the publication date of the document)
Pages：13( 308-320 )