一种结合空间邻域关系特征的面向对象遥感影像变化检测方法

An object oriented remote sensing image change detection method combined with spatial neighborhood relationships

为了充分挖掘遥感影像特征,提高遥感影像变化检测精度,在面向对象遥感图像分割基础上,提出一种遥感影像空间关系特征度量方法,并应用到了变化检测中。首先,通过对两个时期遥感影像叠加分割,提取影像对象;然后,利用目标对象光谱特征及对象与其邻域对象的空间关系特征,构建两个时期影像对象差异特征影像。对象光谱特征参数主要选择对象内像元亮度均值,对象空间关系特征参数主要为目标对象与两个时期的邻域对象之间的像元亮度之间的差值。最后,通过最大期望EM算法自动获取分割阈值,得到影像变化信息。文中利用两期QuikBird影像对实验区域构建多组特征变化矢量影像,实验结果发现,新加入的空间邻域关系特征能较好地提高变化检测精度。

In order to fully exploit the features of remote sensing images and improve the change detection accuracy,based on the object-oriented image segmentation,this paper presents a method for measuring the spatial relationship of remote sensing images and finished related change detection experiments.Firstly,with the composite segmentation of the two-different times remote sensing images,the image objects are extracted.By using the spectral characteristics of target objects and spatial relationships between objects and their neighborhood objects,the image features of two periods are constructed.The mean value of the pixel brightness in the object is selected as the spectral feature parameters,and the feature parameter of the object spatial relationship is mainly the brightness difference between the target object and its neighboring objects in the two periods.Finally,the changed information is automatically obtained with segmentation threshold calculated by the EM method.Two times of QuikBird images were used to construct multiple groups of feature-varying vector images in the experimental area.The comparison showed that the newly added spatial neighborhood features could efficiently improve the change detection accuracy.