Object-oriented and Multi-Feature Hierarchical Change Detection Based on CVA for High-Resolution Remote Sensing Imagery

ZHAO Min,ZHAO Yindi

DOI: https://doi.org/10.11834/jrs.20186293

2018-01-01

Journal of Remote Sensing

Abstract: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 bitemporal 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.