航空科学技术2023，Vol.34Issue(12)：111-117,7.DOI:10.19452/j.issn1007-5453.2023.12.013
面向空中小目标检测任务的YOLOv7改进模型
An Improved YOLOv7 Model for Small Aerial Object Detection
摘要
Abstract
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.关键词
空中目标/目标检测/计算机视觉/深度学习/损失函数Key words
aerial object/object detection/computer vision/deep learning/loss function分类
航空航天引用本文复制引用
董凤禹,魏振忠..面向空中小目标检测任务的YOLOv7改进模型[J].航空科学技术,2023,34(12):111-117,7.基金项目
航空科学基金(201946051002) （201946051002）
国家自然科学基金(52127809) Aeronautical Science Foundation of China(201946051002) （52127809）
National Natural Science Foundation of China(52127809) （52127809）
