YOLOv5与光流相结合的红外小目标检测算法
作者:
作者单位:

作者简介:

通讯作者:

基金项目:


Infrared small target detection algorithm combinedwith YOLOv5 and optical flow
Author:
Affiliation:

Fund Project:

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 资源附件
    摘要:

    针对运动场景下红外小目标轨迹提取不准确的问题,提出采用YOLOv5与光流相结合的方法校正目标轨迹。首先YOLOv5网络,然后对比了LK和HS两种光流算法通过单样本K S检验计算分布拟合度,构建Q Q图得到平台真实运动量,最后结合YOLOv5校正目标轨迹。实验结果表明,LK算法更适合红外图像光流值的提取,YOLOv5与光流相结合的方法在地/空背景下红外图像数据集检测准确率达到90以上,损失率在002以下,对于区分真实和虚假小目标有着重要意义。

    Abstract:

    Aiming at the problem of inaccurate extraction of the trajectory of small infrared targets in sports scenes,a method combining YOLOv5 and optical flow is proposed to correct the trajectory of the target.First,the training network of small infrared targets in YOLOv5 is introduced,and then the LK and HS optical flow algorithms are compared.A single sample K S test is used to calculate the distribution fit;the Q Q diagram is constructed to obtain the true amount of background motion,and finally combined with YOLOv5 to correct the target trajectory.The experimental results show that the LK algorithm is more suitable for the extraction of infrared images optical flow value.The combination of YOLOv5 and optical flow has a detection accuracy of more than 90% and a loss rate of less than 0.02 in infrared image datasets under ground/air background,which is of great significance for distinguishing between real and false small targets.

    参考文献
    相似文献
    引证文献
引用本文

刘宝林,范有臣,秦明宇,谢鹏飞,郭惠超,张来线. YOLOv5与光流相结合的红外小目标检测算法[J].激光与红外,2022,52(3):435~441
LIU Bao-lin, FAN You-chen, QIN Ming-yu, XIE Peng-fei, GUO Hui-chao, ZHANG Lai-xian. Infrared small target detection algorithm combinedwith YOLOv5 and optical flow[J]. LASER & INFRARED,2022,52(3):435~441

复制
分享
文章指标
  • 点击次数:
  • 下载次数:
历史
  • 收稿日期:
  • 最后修改日期:2021-07-25
  • 录用日期:
  • 在线发布日期: 2022-03-22