面向对象的多特征分级CVA遥感影像变化检测

变化矢量分析CVA方法在中低分辨率遥感影像变化检测中已得到广泛应用,但由于高分辨率遥感影像存在不同地物尺度差异大、不同类别地物光谱相互重叠的问题,因此对于高分影像的变化检测具有局限性。为提高高分影像变化检测精度,提出了一种面向对象的多特征分级CVA变化检测方法,首先,利用基于区域邻接图的影像分割方法分别对两时相遥感影像进行多尺度分割,提取分割图斑的光谱、纹理和形状特征;然后,在各级尺度下,分别运用随机森林方法进行特征选择,计算CVA变化强度图;最后,根据信息熵对多级变化强度图进行自适应融合,利用Otsu阈值法检测变化区域,并与仅考虑光谱特征的分级CVA变化检测方法、像元级多特征CVA变化检测方法以及仅考虑光谱特征的像元级CVA变化检测方法进行比较分析。实验表明:与比较方法相比,本文方法的变化检测精度较高,误检率和漏检率较低。

Object-oriented and multi-feature hierarchical change detection based on CVA for high-resolution remote sensing imagery

With increasing image resolution, change detection for high-resolution images has become one of the most important aspects of remote sensing research. Change Vector Analysis (CVA) is an effective method that has been widely used in change detection for low- or moderate-resolution remote sensing images. However, processing high-resolution data involves limitations caused by spectral heterogeneity and objects with different scales. Thus, CVA must be combined with object-oriented methods. However, the performance of most object-oriented methods depends on the results of image segmentation, which are unstable due to the difficulty in determining the optimum scale. Taking advantage of the rich spatial information in high-resolution images is obviously important. Considering the aforementioned problems, this work proposes an object-oriented and multi-feature hierarchical change detection method based on CVA.Bi-temporal high-resolution remote sensing images are hierarchically segmented. Hierarchical image segmentation is realized with an image segmentation method based on a region adjacency graph. A logical OR is then applied to corresponding segmentation levels of the bi-temporal images. On the basis of spatial characteristics, spectral, texture, and shape features are extracted. A gray level co-occurrence matrix is used as a texture feature, and a geometric moment is used as a shape feature. Feature selection is realized with a random forest algorithm. Then, CVA is conducted to calculate the hierarchical magnitude images according to the optimal feature vectors. The final change magnitude image is obtained by fusing the hierarchical magnitude images using an adaptive fusion method. Finally, the Otsu algorithm is used to determine the change threshold values and thereby realize change detection.The change detection result of the object-oriented and multi-feature hierarchical CVA method is compared with those of the hierarchical CVA that only uses spectral features, the multi-feature pixel-based CVA, and the pixel-based CVA that only uses spectral features. Experiment outcomes show that the proposed method offers higher change detection accuracy and greater stability than the other three methods. It also achieves reduced false alarm rate and missing alarm rate in the change detection result. The proposed method can also reduce salt-and-pepper noise and produce entire changed objects.The unit for analysis in the proposed object-oriented and multi-feature hierarchical change detection based on CVA is an image object at different scales. The changes are detected at different hierarchy levels to adapt to the different characteristics of objects with different scales and to avoid the difficulty in determining the optimal segmentation scale. Thus, the impact of segmentation on change detection precision is minimized. In addition, multiple feature extraction and optimal feature selection ensure that the spectral and spatial information of high-resolution images is fully utilized while the characteristic redundancy is reduced. Therefore, the influences of complex spectral characters are avoided. The proposed method demonstrates high performance and high reliability.