Abstract: Aiming at the problem that SLIC segmentation algorithm does not consider elevation information in remote sensing image segmentation, which leads to poor segmentation effect of some features, this paper proposes the SLIC super-pixel segmentation of remote sensing image fused with elevation information and the double-threshold region merging method based on elevation grading. Firstly, the elevation information is introduced into the initial clustering segmentation threshold to obtain the initial segmentation result with dependence on both spectral gradient and elevation gradient. Then, the data structure of neighborhood array is adopted on the basis of pre-segmentation to establish the similarity metric by weighting the spectral information of different regions combined with elevation information. Finally, the graded elevation threshold is set, and different merging threshold weights are set according to the elevation difference between different regions perform region merging. The proposed method is validated by using the low-altitude remote sensing image data with fused elevation information and the dataset provided by the international society for photogrammetry and remote sensing, and the results show that good segmentation results are achieved by introducing elevation information in the super-pixel segmentation and region merging method based on spectral information.

Key words: remote sensing images, image segmentation, SLIC super-pixel, region merging, neighborhood array