HIERARCHICAL MULTINOMIAL LATENT MODEL WITH G0 DISTRIBUTION FOR REMOTE SENSING IMAGE SEMANTIC SEGMENTATION

Considering the scattering statistics and multi scale characteristics of the remote sensing images, this paper presents a hierarchical multinomial latent model with G(0) distribution (HML-G(0)) for remote sensing image semantic segmentation. In the proposed approach, hierarchical multinomial latent model is proposed to capture the multi scale information of the remote sensing images. Moreover, the flexibility of G(0) distribution is plugged into the hierarchical multinomial latent model for the segmentation of various types of land covers. Then, the developed Bayesian inference on the quadtree is incorporated in our approach, and the semantic segmentation map is achieved by bottom-up and top-down probability computation. Experimental results demonstrate that our proposed hierarchical scheme produces the semantic segmentation maps, and the exhibiting performance improvements in terms of labeling consistency and the detail preservation.