Research On Unsupervised Classification Of SAR Sea Ice Based On High-Order Neighborhood MRF

The SAR classification algorithm sea ice random field image is affected and interfered by random factors in many aspects during its formation.The sea ice image is interfered and affected by random factors such as incidence angle,frequency,polarization mode,coherent spots,climate and other factors,Which brings great challenges to the correct classification and processing of sea ice random field images.Existing sea ice image classification and processing algorithms based on random field model classification methods can easily lead to excessive smoothing and misclassification if complex neighborhood and context random field models are introduced,so this paper mainly proposes a The hybrid high-order neighborhood MRF classification algorithm improves the neighborhood clustering method and uses the neighborhood and context information in the Markov random field image model algorithm to improve the classification accuracy.The main academic research directions and content of this article are as follows:First of all,to deal with the structural problem of uneven sea ice gray scale in a single area,the seawater thin ice area with a rough surface and the surrounding seawater rich ice area are highly overlapping,and the boundary is blurred,which is difficult to distinguish accurately.In this paper,by analyzing and calculating the joint probability distribution of the area,the single and complex area is divided into two homogeneous and heterogeneous areas.For the single and homogeneous area,there is no obvious complex pixel structure,so the center The pixels of a region and all the complex pixels in its neighborhood should be closely related.They have a great possibility and opportunity to be classified in the same region.For heterogeneous characteristic regions,there are obvious changes in the material structure in the region,and light and dark homogeneous regions alternately appear in the heterogeneous characteristic regions.The adaptive combined region mixed Gaussian radial basis function and the combined Gaussian core basis function combined with ridgelet function are designed for the light and dark homogeneous and heterogeneous characteristic regions respectively,which represents the sea ice isotropy and anisotropy The interrelationship is used to classify the characteristics of different regions and the classification of riSAR sea ice in heterogeneous regions.Then,in view of the uneven brightness in thick ice areas,the phenomenon of excessive brightness appears in the local regional evidence space.In order to make better use of the neighborhood to polarize the context information and effectively adjust the correlation of the transfer evidence space,the traditional low-order neighborhood model limits the time and complexity of the higher-order MRF model to capture complex context transfer evidence information This paper extends the original neighborhood based on the MRF evidence model to a higher-order MRF model and builds a space for transferring evidence.This paper first calculates the likelihood function and space of the thick ice bright area likelihood and over bright area when constructing the space of the thick ice transfer over bright area evidence,and then calculates the space method of thick ice transfer evidence space defined in all directions Get space for evidence of thick ice transferring across bright areas.Finally,comparative analysis experiments using traditional unsupervised classification algorithms such as classic MRF,k-means,Iso-Data,and FCM prove that the classification results obtained by the improved high-order MRF are better than those obtained by other classification algorithms.The details are maintained and the classification accuracy is improved,proving the effectiveness of the algorithm in this paper.