摘要
传统的遥感图像变化检测方法未能充分利用像素上下文信息,导致精度较低。针对这一问题,提出一种不需要像素概率分布假设、无监督的上下文光谱角映射图像变化检测方法。在像素变化的判别测试中,利用空间上下文信息特征提高了变化检测精确度,此方法可用于卫星遥感图像中,利用上下文光谱角映射创建相似图像,基于K-均值聚类算法将其分为有变化和无变化两类,以此生成映射图像。通过定性和定量分析,将实验结果与最大似然估计法(MLC)结果相比较。研究表明,无监督上下文光谱角映射图像变化检测方法精确度更高,可用于二时刻图像和多光谱图像变化检测问题。
Traditional remote sensing image change detection methods didn' t fully utilize context-sensitive information, so that the accuracy was low. To deal with this problem, this paper proposed an unsupervised context-sensitive spectral angle mapper based image change detection method,which didn't need probability distribution assumption of pixels. On the criterion of pixels change test,it used spatial context information, so as to improve the change detection accuracy. This method could be used in the satellite remote sensing images. It created an image of the similar spectral angle mapping by using context. Based on Kmeans clustering algorithm,it divided the mapping into two kinds of changed and unchanged. Through the qualitative and quantitative analysis,it compared the test results and the MLC results. Research shows that the accuracy of unsupervised image change detection method based on context spectral angle mapping is higher. It can be used in hi-temporal image and muhispectral image change detection.
作者
高雷阜
李超
Gao Leifu Li Chao(College of Science, Liaoning Technical University, Fuxin Liaoning 123000, China)
出处
《计算机应用研究》
CSCD
北大核心
2016年第12期3889-3891,3896,共4页
Application Research of Computers
基金
国家自然科学基金资助项目(11401284)
高校博士学科点专项科研联合基金资助项目(20132121110009)
关键词
遥感图像
无监督
上下文
变化检测
K-均值聚类
remote sensing image
unsupervised
context-sensitive
change detection
K-means clustering