Remote sensing image segmentation based on self-organizing map at multiple-scale

摘要

This paper proposes a segmentation method based on K-mean and SOM network. Firstly remote sensing image is decomposed by wavelet transform at multiple-scale. Secondly the directional eigenvector of the image is constructed based on the wavelet transform. At coarser scale, we construct 4-dimension eigenvector with feature images, and the images are roughly segmented by K-means algorithm. Then we construct 4-dimension eigenvector with other feature images at fine scale. Based on the results in K-means segmentation and the eigenvector of remote-sensing images at fine scale the images are segmented by SOM network. The experiments about the images segmentation are done in two different ways, one of which is K-means and SOM network simultaneously, and the other of which is mere K-mean. The experiments show that the former has better segmentation results and higher efficiency.