摘要
在遥感图像变化检测问题中,半监督的支撑向量机分类法由于可以充分利用原始遥感图像中所有波段的信息,有利于变化信息的检测.但是在实际问题中,当变化区域面积占整幅图像的比例相对较大或较小时,这种方法并不能准确地检测出变化信息.针对这一问题,提出一种基于分割窗的遥感图像变化检测方法.该方法结合半监督的支撑向量机分类法,先将差异图像分割成子图像,通过子图像的最优超平面来确定差异图像的分类超平面.实验结果表明,该方法能较好地解决遥感图像中变化区域相对较大或较小时半监督支撑向量机分类法不能准确检测的问题,具有很好的变化检测性能.
For change detection of remotely sensed images, the Semi-supervised SVM method results in the high accuracy of change detection because it can sufficiently exploit information of all bands of multichannel remotely sensed images. However, in practice this method cannot detect the change information accurately when percentage of changed area in whole sensed image is relatively small or large. In order to solve this problem, a split window-based method is proposed to improve the Semi-supervised SVM. It firstly splits difference image into a ~t of subqmages, and then determine the classification hyper- planes of the difference image with the optimal hy-per-plane of the sub-image Experimental results demonstrate that the proposed method can detect change information accurately even if percentage of changed area in whole sensed image is relatively small or large, and improve detection accuracy obviously relative to the traditional semi-supervised SVM method.
出处
《复旦学报(自然科学版)》
CAS
CSCD
北大核心
2010年第2期190-196,共7页
Journal of Fudan University:Natural Science
基金
国家自然科学基金(60672116)
国家高技术研究发展计划("863"计划)(2009AA12Z115)资助项目
关键词
变化检测
分割窗
半监督支撑向量机
遥感图像
change detection
split window
semi supervised SVM
remotely sensed images