奇异值分解和阈值分割的遥感自适应变化检测

Adaptive change detection algorithm of remote sensing images based on SVD and OTSU

针对传统遥感变化检测算法中差值影像构造方法不容易提取弱变化信息,以及变化阈值需要人工干预的不足,提出一种基于奇异值分解(SVD)和最大类间方差法(OTSU)阈值分割的遥感影像变化检测算法。首先计算两期遥感影像的多波段差值影像,对其进行矢量化后所构造矩阵进行奇异值分解,并计算差值影像的变化强度,选取奇异值分解主分量投影和变化强度作为表征两期影像变化的特征量;然后通过最大类间方差法对上述两个特征量进行阈值分割,得到变化检测结果;最后通过对比实验以及精度验证,证明了该变化检测方法相比传统方法能够更精确地提取出变化信息,而且能够自适应获取分割阈值。

In traditional remote sensing change detection methods,difference image construct methods could not detect weak change information accurately,and the change detection segmentation threshold could not be obtained intelligently.For these problems,a modified change detection method based on singular value decomposition（SVD）and OTSU threshold segmentation was discussed in this paper.Firstly,difference image of two temporal remote sensing images was calculated,singular value was decomposed by constructing of vector matrix of difference image,change intensity was calculated,and results of SVD principle component and change intensity were taken as main features of change detection.Secondly,segmentation thresholds were discriminated by OTSU segmentation method,and results of change detection could be obtained objectively.Finally,by comparison experiment and accuracy analysis,it illustrated that this modified method could be prior to traditional methods,it could more accurately extract change information,it could obtain segmentation threshold adaptively,and results of change detection were more objective and dependable.