摘要
提出一种基于主题模型的高分辨率遥感影像变化检测方法。将前后两期遥感影像对应的像素点对作为基本单位,提取其邻域亮度相关度、均值、标准差以及邻域回归直线的斜率、截距等低层次特征,在此基础上映射得到像素点对的高层次视觉单词特征,并通过潜在狄利克雷分配模型进行分析,挖掘其潜在的主题信息,即变化与不变,从而实现变化检测。实验结果表明,该方法能够有效检测高分辨率遥感影像的变化。
A novel change detection method based on topic model is proposed for high resolution remote sensing images. It takes every pixel pairs of the bi-temporal remote sensing images as a basic unit, and extracts their low-level features, such as the relevacy, mean value, standard deviation of neighbor brightness, the slope and intercept of neighbor regression line. Then the high-level visual words mapped by theses low-level features are generated. After that, by utilizing the classical topic model of Latent Dirichlet Allocation（LDA） to analyze, the latent topic information is to be found, that is, changed or unchanged, thereby the goal of change detection is achieved. Experimental results show that this method can effectively detect changes in high resolution remote sensing images.
出处
《计算机工程》
CAS
CSCD
2012年第15期204-207,共4页
Computer Engineering
基金
国家"973"计划基金资助项目(2006CB701303)
国家自然科学基金资助项目(41071256)
教育部高等学校博士点基金资助项目(20090073110018)
关键词
主题模型
视觉单词
高分辨率
潜在狄利克雷分配
遥感影像
变化检测
topic model
visual word
high resolution
Latent Dirichlet Allocation（LDA）
remote sensing image
change detection
