摘要
针对遥感飞机图像中存在背景复杂、密集小目标及目标遮挡等导致飞机检测精度不高的问题,提出一种基于改进YOLOv5s的遥感图像飞机检测算法。采用稠密连接增强特征传播与特征重用,减轻训练过程中的梯度消失问题;引入注意力机制进行自适应特征细化,提升密集小飞机目标的检测性能;改进损失函数增强目标被遮挡或多目标重叠情况下的检测效果。实验结果表明,改进后的算法能显著提高遥感图像飞机检测的精度,具有较强的实用性与适用性。
Aiming at the problem of low accuracy of remote sensing image aircraft detection due to complex background,dense small targets and occluded targets,a remote sensing image aircraft detection algorithm based on improved YOLOv5s was proposed.The dense connections were used to enhance feature propagation and feature reuse,so as to alleviate the problem of gradient disappearance in the training process.The attentional mechanism was introduced for adaptive feature refinement to improve the detection performance of dense small targets detection.The loss function was improved to enhance the detection effect under occlusion and multi-target overlap.According to the experimental results,the optimized algorithm can availably improves the accuracy of remote sensing image aircraft detection,and has strong practicability and applicability.
作者
鄢奉习
徐银霞
蔡思远
祁泽政
YAN Feng-xi;XU Yin-xia;CAI Si-yuan;QI Ze-zheng(Hubei Key Laboratory of Intelligent Robot,Wuhan Institute of Technology,Wuhan 430205,China;School of Computer Science and Engineering,Wuhan Institute of Technology,Wuhan 430205,China)
出处
《计算机工程与设计》
北大核心
2023年第9期2794-2802,共9页
Computer Engineering and Design
基金
国家自然科学基金项目(62171327)
中国高校产学研创新基金-新一代信息技术创新基金项目(2020ITA07015)
武汉工程大学研究生教育创新基金项目(CX2021267)。
关键词
YOLOv5s
深度学习
目标检测
人工智能
遥感图像
飞机识别
计算机视觉
YOLOv5s
deep learning
target detection
artificial intelligence
remote sensing image
aircraft recognition
computer vision
