Composite Classification Of SAR Sea Ice Images Based On Hierarchical CRF

With the diversification of remote sensing images,the shortcomings of classification based on single resolution data have gradually emerged.Low resolution remote sensing images generally have the problems of poor resolution and low classification accuracy of mixed pixels.However,the coverage of high-resolution remote sensing images is generally small and the classification processing time is long.The core idea of the composite classification method is to use high-resolution images to guide the classification of low-resolution images.It can not only ensure the classification accuracy to a certain extent,but also effectively solve the problem of small coverage.Based on several SAR sea ice images with different resolutions,this paper focuses on the research of SAR sea ice image composite classification based on hierarchical conditional random field(CRF)algorithm.The specific work of this paper are as follows:Firstly,in view of the large difference in SAR sea ice characteristics in composite classification,this paper selects the most suitable characteristics for sea ice composite classification through experiments.For SAR sea ice images with high resolution difference,due to different data sources and polarization modes,eight texture features are extracted for feature selection.By comparing Kappa coefficient,accuracy rate,recall rate and FIA Formula 1 World Championship value of classification results,Gabor texture features that are most suitable for sea ice composite classification are selected.Secondly,a conditional random field algorithm combined with hierarchical method is proposed.Conditional random field algorithm can capture the semantic association between multiple tags.Because of this association,it can minimize the influence of texture,color,location and edge differences of images with different resolutions,thus ensuring the classification accuracy.However,due to the large amount of SAR image data,it is difficult to guarantee the classification speed by using only conditional random field algorithm.Therefore,this paper proposes a conditional random field algorithm combined with hierarchical method.This algorithm reduces the classification time and ensures the classification accuracy.It is more suitable for SAR sea ice image composite classification in this paper.Finally,the classification results of the proposed algorithm and three traditional composite classification methods are compared and analyzed.For composite classification,how to verify the validity of classification is a difficult problem.This paper designs a set of methods that can effectively verify the effect of composite classification.Based on this method,the comparison of Kappa coefficients proves the effectiveness of the proposed algorithm.