摘要
针对空中小目标检测任务,本文进行了基于改进的YOLOv7模型的空中小目标检测算法的研究,解决了当前空中小目标检测算法推理速度慢和检测精度不足的问题。首先,建立了涵盖多种目标尺度、姿态和天气条件的飞机目标基准数据集;其次,在YOLOv7基准模型的基础上,提出了一种基于广义特征金字塔网络和Wasserstein度量的目标检测方法;最后,在公开数据集与自建数据集上对该方法与其他主流算法进行了对比试验,结果表明,相较于原始YOLOv7模型,改进模型在自建数据集上对小目标检测的平均精确率提高了7.3%,并且推理速度高于大部分主流检测算法。本文研究为空中小目标检测任务提供了一种快速且高精度的检测算法,对于相关算法在航空航天领域的进一步工程应用具有重要的推动作用。
Addressing the challenges of slow inference speed and limited detection accuracy in small airborne target detection tasks,this paper investigated an enhanced detection algorithm grounded in the YOLOv7 model.Firstly,an encompassing benchmark dataset for aircraft targets was established,encapsulating varying scales,orientations,and weather conditions.Secondly,a novel detection approach was introduced,leveraging a generalized feature pyramid network and Wasserstein distance metric within the YOLOv7 framework.Finally,comparative evaluations encompassing public and self-constructed datasets validated the method against mainstream algorithms.The enhanced model exhibited a 7.3%boost in small target detection precision on self-constructed datasets,alongside accelerated inference speed surpassing mainstream counterparts.This research delivers a swift and highly precise detection algorithm tailored for small airborne target detection,contributing significantly to advancing the application of related algorithms in aerospace engineering.
作者
董凤禹
魏振忠
Dong Fengyu;Wei Zhenzhong(Beihang University,Beijing 100191,China)
出处
《航空科学技术》
2023年第12期111-117,共7页
Aeronautical Science & Technology
基金
航空科学基金(201946051002)
国家自然科学基金(52127809)。
