低空无人机实时探测的PBE-YOLOv8n算法
PBE-YOLOv8n Algorithm for Real-Time Detection of a Low Altitude UAV
郭汝昂 1任帅 1张航1
作者信息
- 1. 西安石油大学计算机学院,西安,710065
- 折叠
摘要
面向低空无人机"黑飞""滥飞"带来的威胁,针对现存低空无人机目标检测算法检测精度较低、实时性较差的问题,本文提出一种基于YOLOv8n改进的PBE-YOLOv8n低空无人机目标检测模型.使用局部卷积(PConv)代替普通卷积(Conv),设计全新的快速跨阶段局部层卷积(C2f_Faster)模块代替跨阶段局部层卷积(C2f)以实现模型轻量化;使用黄金集散(Gold-YOLO)结构替换颈部路径聚合网络(PANet)结构,保留更多渐层特征,提高检测精确性;在颈部网络中引入高效多尺度注意力(EMA)机制,捕捉局部重要信息,以提高模型的特征融合能力;使用智能交并比(WIoU)边界损失函数代替原损失函数,提升网络的边界框回归性能.实验结果表明,本文提出的PBE-YOLOv8n算法在精确度和速度上都有所提升,证明了该改进算法的有效性和先进性.
Abstract
An improved PBE-YOLOv8n low-altitude UAV target detection model is proposed to address the issues of low detection accuracy and poor real-time performance in existing low-altitude UAV target detec-tion algorithms for the threats posed by unauthorized and uncontrolled drone activities("black-fly"and"illegal-fly").The model is based on YOLOv8n with modifications.PBE-YOLOv8n incorporates partial convolution PConv instead of common Conv layers to achieve model light weighting.A new CSPLayer_2Conv_faster C2f_Faster module is designed to replace part of CSPLayer_2Conv C2f for improved real-time performance.The Path Aggregation Network(PANet)structure in the neck part is replaced with Gold-YOLO to preserve more gradient features and enhance detection accuracy.Efficient Multi-Scale Attention(EMA)is introduced in the neck network to capture local important information for improved feature fusion capability.The original loss function is replaced with the Wise Intersection over Union(WIoU)boundary loss function to enhance the network's bounding box regression performance.Experimental results demonstrate that the proposed PBE-YOLOv8n algorithm achieves improvements in both accuracy and speed,validating the effectiveness and ad-vancement of the improved algorithm.
关键词
YOLOv8/低空无人机/快速跨阶段局部层卷积/黄金集散/高效多尺度注意力/智能交并比Key words
YOLOv8/low altitude UAV/C2f_Faster/Gold-YOLO/Efficient Multi-Scale Attention/Wise Intersection over Union引用本文复制引用
出版年
2024
NETL
NSTL
万方数据