欢迎访问中国科学院大学学报,今天是

中国科学院大学学报 ›› 2022, Vol. 39 ›› Issue (1): 102-109.DOI: 10.7523/j.ucas.2020.0005

• 电子信息与计算机科学 • 上一篇    

一种联合光谱-对象-时间特征的遥感影像变化检测方法

曹州1,2, 刘士彬1, 马勇1, 姚武韬1,2, 姜丽媛1,2   

  1. 1. 中国科学院空天信息创新研究院, 北京 100094;
    2. 中国科学院大学, 北京 100049
  • 收稿日期:2020-01-08 修回日期:2020-04-16 发布日期:2021-05-31
  • 通讯作者: 刘士彬
  • 基金资助:
    国家科技基础资源调查专项(2018FY100500)资助

A change detection method by combining spectral-object-temporal features for remote sensing imagery

CAO Zhou1,2, LIU Shibin1, MA Yong1, YAO Wutao1,2, JIANG Liyuan1,2   

  1. 1. Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing 100094, China;
    2. University of Chinese Academy of Sciences, Beijing 100049, China
  • Received:2020-01-08 Revised:2020-04-16 Published:2021-05-31

摘要: 针对在遥感大数据时代背景下,传统变化检测方法的精度和自动化程度难以满足实际应用需求,提出一种联合光谱特征、对象特征和时间特征的遥感影像变化检测方法。在提取遥感影像多种特征的基础上,利用双向长短期记忆网络,提取光谱-对象-时间特征,实现双时相影像变化信息的有效提取。基于双时相中分辨率遥感影像的实验结果表明:本方法的总体精度超过0.9,Kappa系数达到0.84。相较于传统的变化检测方法,可以有效提高变化检测的精度和自动化程度。

关键词: 变化检测, 光谱-对象-时间特征, 双向长短期记忆网络

Abstract: According to the fact that traditional change detection methods are difficult to meet the needs of practical applications in the era of big data of remote sensing, this paper proposes a method by combining spectral-object-temporal features to solve this problem. On the basis of extracting various features of remote sensing images, the Bi-LSTM network is used to extract the joint spectrum-temporal-object feature to obtain the effective information of changes in bi-temporal images. Experimental results based on bi-temporal medium-resolution remote sensing images show that the overall accuracy of this method is greater than 0.9, and the Kappa coefficient reaches 0.84. Compared to traditional methods, the proposed method effectively improves the accuracy and degree of automation of change detection.

Key words: change detection, spectral-object-temporal feature, Bi-LSTM

中图分类号: