Change Detection in Multispectral Remote Sensing Images Based on Optimized Fusion of Subspaces

摘要

In this paper, an effective approach is proposed for unsupervised change detection in multispectral remote sensing images. Firstly, the spectral-spatial information joint distribution of multispectral remote sensing images is achieved by multiscale morphological tools. Thus more geometrical details of images are extracted while exploiting the connections of a pixel and its adjacent regions. Subsequently, the difference images of change vector analysis and spectral angle mapper are generated according to the difference of spectral vectors magnitude and direction, respectively. Finally, the two difference images are combined by optimized fusion algorithm named affinity aggregation based on Nytrom spectral clustering to obtain the binary change mask. Experimental results show that the proposed method not only detects weak changes but also effectively maintains the integral geometry of objects.