摘要
为了解决在动态飞机监管系统中,遥感图像飞机多目标检测准确率低的问题,提出一种目标检测方法.首先引入两种全新的旋转不变性特征,中心质点角与H向量;接着对遥感飞机图像进行滑窗检测,分别计算每一重叠块的中心质点角集、H向量相关系数;根据模板特征匹配度设计了相应的得分系统,再结合非极大值抑制等算法确定检测窗口内是否存在飞机.用不同场景下的遥感图像飞机进行多目标检测实验,结果表明该方法平均F1分数达到90%以上,相比传统方法,召回率与查准率更高,且适用范围更为广泛.
To improve the accuracy of the multi-target detection for remote sensing image in the dynamic aircraft supervision system,a target detection method was proposed.First,two new rotation-invariance features,named center-particle angle and H-vector,are introduced.Then,the sliding detection window is used to calculate the center-particle angle and correlation coefficient of H-vector.Also,the corresponding scoring system is designed according to the matching degree of template feature to determine if there is a plane in the detection window under the assistant of non-maximum suppression.The remote sensing images of aircraft under different scenes were detected in experiment,the results show that the average F1-score reaches above 90%,and both of the recall rate and precision are higher than some traditional methods in wider scope of application.
作者
林亿
赵明
潘胜达
安博文
LIN Yi;ZHAO Ming;PAN Sheng-da;AN Bo-wen(College of Information Engineering,Shanghai Maritime University,Shanghai 201306,China)
出处
《光子学报》
EI
CAS
CSCD
北大核心
2019年第6期153-162,共10页
Acta Photonica Sinica
基金
国家自然科学基金(Nos.61302132,61504078,41701523)
上海市教育发展基金会"晨光计划"(No.13CG51)
广西省教育厅基金(No.YB2014207)~~
关键词
模式识别
目标检测
特征匹配
遥感图像
飞机监管
Pattern recognition
Target detection
Feature matching
Remote sensing images
Aircraft supervision
