Segmentation Of SAR Sea Ice Image Based On Region Splitting And Adaptive Refinement Process

SAR sea ice image segmentation is an important application of remote sensing system and an important part of the sea ice monitoring. It is indispensable in the understanding of the arctic climate system and offering the ship navigation safety in the waters area which may form to be sea ice. Unsupervised segmentation method usually is applied in the SAR sea ice image segmentation for the limited of training samples., The most commonly used model is the Markov Random Field (MRF) in the SAR sea ice image segmentation. The MRF method treat the nature of SAR speckle noise by a statistically optimal mode, meanwhile, providing a effective spatial context model for regularization. However, in the traditional MRF method, the models of feature and spatial context are always stationary, and their parameters are for the global estimation. For the incident and environmental factors, such models are restricted in nonstationary SAR sea ice condition, those resulting in a poor segmentation accuracy in a complex scene of SAR sea ice.In order to improve the segmentation accuracy of complex SAR sea ice image, this paper proposes an image segmentation method based on region splitting and binary tree hierarchical structure adaptive adjustment. Firstly, the global iterative weights based on MRF model is used to complete the initial region merging, and the merging process is described as a form of binary tree. In this proposed hierarchical clustering algorithm, there is a positive correlation between the scale of the object in scene and the number of binary tree nodes. The subsequent refinement of region splitting does not generate new regions but only rolls back to a previous configuration. The scale weights of spatial context model is adjusted adaptively, according to the complexity of objects of different regions in the scene. And the updated weights renew the regional merging. The experiment results show that this method effectively improves the segmentation accuracy of SAR sea ice complex scenes image.