Multi-granularity synthesis segmentation for high spatial resolution Remote sensing images

35th International Symposium on Remote Sensing of Environment

地球科学;生态环境科学

Yi, Lina^1 ; Liu, Pengfei^1 ; Qiao, Xiaojun^1 ; Zhang, Xiaoning^1 ; Gao, Yuan^1 ; Feng, Boyan^1

College of GeoScience and Surveying Engineering, China University of Mining and Technology, Beijing 100083, China^1

关键词: High spatial resolution;    Landuse classifications;    Multi-scale image segmentation;    Multiscale segmentation;    Mumford Shah function;    Remote sensing images;    Segmentation accuracy;    Supervised segmentation;   

Others  :  https://iopscience.iop.org/article/10.1088/1755-1315/17/1/012197/pdf
DOI  :  10.1088/1755-1315/17/1/012197

学科分类：环境科学（综合）

来源: IOP

【 摘 要 】

Traditional segmentation method can only partition an image in a single granularity space, with segmentation accuracy limited to the single granularity space. This paper proposes a multi-granularity synthesis segmentation method for high spatial resolution remote sensing images based on a quotient space model. Firstly, we divide the whole image area into multiple granules (regions), each region is consisted of ground objects that have similar optimal segmentation scale, and then select and synthesize the sub-optimal segmentations of each region to get the final segmentation result. To validate this method, the land cover category map is used to guide the scale synthesis of multi-scale image segmentations for Quickbird image land use classification. Firstly, the image is coarsely divided into multiple regions, each region belongs to a certain land cover category. Then multi-scale segmentation results are generated by the Mumford-Shah function based region merging method. For each land cover category, the optimal segmentation scale is selected by the supervised segmentation accuracy assessment method. Finally, the optimal scales of segmentation results are synthesized under the guide of land cover category. Experiments show that the multi-granularity synthesis segmentation can produce more accurate segmentation than that of a single granularity space and benefit the classification.