基于YOLO的无人机小目标高精度检测改进模型OA
An improved high-precision model for UAV small object detection based on YOLO
无人机图像目标检测作为多领域研究的基础技术,被广泛应用于各类复杂场景.然而,航拍图像中目标尺寸小、分布密集、背景复杂,导致现有检测模型在漏检、误检和定位精度方面存在不足.为提升检测性能,提出一种基于YOLO的无人机小目标高精度检测模型.首先,提出选择性卷积模块(Selective Convolution Block,SCB),用于替代原跨阶段结构,降低计算复杂度;其次设计了小目标尺度序列融合结构(Small Target Scale Sequence Fusion,STSSF),以增强多尺度特征融合能力;最后提出共享卷积检测头(Shared Convolution Precision Detection,SCPD),提升特征提取效率和跨尺度特征的一致性.在VisDrone2019-DET数 据 集 上,YOLO-LiteMax在mAP@0.5指标上达到45.2%,较YOLOv8提升5.9%;在APsmall 指标上提升3.8%,展现出优异的小目标检测精度与实用性.
As a fundamental technology for multi-domain research,target detection in unmanned aerial vehicle(UAV)imagery has been widely applied to various complex scenarios.However,aerial images of-ten contain small,densely distributed objects against complex backgrounds,leading to challenges such as missed detections,false positives,and inaccurate localization in existing models.To improve detec-tion performance,this paper proposes a YOLO-based high-precision detection model for small UAV ob-jects.First,a selective convolution block(SCB)is adopted to replace the original cross-stage module,reducing computational complexity.Second,a small target scale sequence fusion(STSSF)structure is designed to enhance multi-scale feature fusion.Third,a shared convolution precision detection(SCPD)module is proposed to improve feature extraction efficiency and cross-scale feature consistency.Experi-mental results on the VisDrone2019-DET dataset show that YOLO-LiteMax achieves a mAP@0.5 of 45.2%,which is 5.9%higher than that of YOLOv8,along with a 3.8%increase in APsmall.These re-sults demonstrate superior precision and practical applicability in small-object detection.
杨畅;苏健;张健
南京信息工程大学 计算机学院,江苏 南京 210044南京信息工程大学 软件学院,江苏 南京 210044南京信息工程大学 计算机学院,江苏 南京 210044
信息技术与安全科学
无人机目标检测特征提取图像处理
unmanned aerial vehicle(UAV)object detectionfeature extractionimage processing
《南京邮电大学学报（自然科学版）》 2026 (3)
132-139,8
国家自然科学基金(61802196)资助项目
评论