基于改进YOLOv8s的无人机航拍图像小目标检测算法  

Small Target Detection Algorithm for UAV Aerial Photography Images Based on Improved YOLOv8s

作　　者：方伟 张亚[1] FANG Wei;ZHANG Ya(School of Artificial Intelligence,Anhui University of Science and Technology,Huainan Anhui 232001,China)

机构地区：[1]安徽理工大学人工智能学院,安徽淮南232001

出　　处：《兰州工业学院学报》2025年第1期82-88,共7页Journal of Lanzhou Institute of Technology

基　　金：安徽省高校自然科学基金(2022AH50801)。

摘　　要：针对无人机航拍图像小目标检测精度低的问题,提出一种改进YOLOv8s的小目标检测算法。算法网络中,通过增加协调注意力机制(Coordinate Attention,CA)来提升目标特征提取能力;改进C2f模块并在颈部网络中引入CARAFE上采样算子,从而降低模型参数,增强多尺度特征融合能力。同时,添加一个160×160尺度的小目标检测层去替换大目标检测层,减少网络深度,提高模型对小目标的敏感度,解决误检漏检问题。实验结果表明:在VisDrone2019数据集上,相较于原YOLOv8s,改进算法的精确率P提升了6.95%,召回率R提升了8.36%,平均精度mAP50提升了13.06%,并且参数量降低了73.02%。改进算法在小目标检测方面展现出良好性能,在多项指标上优于其他YOLO系列算法,体现了其在小目标检测领域的竞争力。Aiming at the problem of low precision in small target detection in UAV images,an improved small target detection algorithm based on YOLOv8s is proposed.In this algorithm,a Coordinate Attention(CA)mechanism is introduced to enhance the capability of target feature extraction.Furthermore,improvements are made to the C2f module,and the CARAFE upsampling operator is incorporated into the neck network to reduce model parameters and enhance multi-scale feature fusion.Additionally,a 160×160 scale small target detection layer is added to replace the large target detection layer,reducing network depth,increasing sensitivity to small targets,and mitigating false positives and false negatives.Experimental results on the VisDrone2019 dataset demonstrate that compared to the original YOLOv8s,the proposed algorithm achieves a 6.95%increase in precision(P),an 8.36%increase in recall(R),and a 13.06%improvement in average precision(mAP50),with a reduction of 73.02%in parameters.The improved algorithm demonstrates excellent performance in small object detection,outperforming other YOLO series algorithms across multiple metrics,highlighting its competitiveness in the field of small object detection.

关 键 词：YOLOv8 航拍图像 目标检测 注意力机制 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]