Arctic Sea Ice SAR Image Classification Based On MRF Model

The Arctic sea ice is rapidly decreasing with the global warming,so that the openning of new Arctic routes become possible,which is conducive to China's international trade and maritime transport development,it also provide opportunities for China to get rid of dependence on single energy channel and enhance national energy security.Therefore,the research on Arctic sea ice has important economic value and strategic significance.Synthetic aperture radar(SAR)has the advantages of all-day,all-weather,high resolution,and has become an important meathod of sea ice monitoring in recent years.This paper focuses on the key technologies of SAR sea ice images acquisition and interpretation,such as SAR imaging algorithm,sea ice feature extraction and MRF model classification.The main work of this paper is as follows:(1)This paper studies the Micro SAR system hardware composition,design block diagram and technical indicators,The Micro SAR is a frequency modulated continuous-wave synthetic aperture radar(FM-CW SAR),and it is equipped on small-scale unmanned aerial vehicles(UAVs),it has the advantages of low power consumption,low cost and high resolution.Due to the antenna Azimuth changes greatly,improves the traditional imaging Algorithms,a Azimuth filter of relation with distance is introduced to imaging Algorithms to reduce the impact of the rapid disturbance of the SAR platform,And images are multi-looks processed to suppress speckle noise,compared with the unimproved imaging results,the improved algorithm has better imaging quality.(2)A recognition algorithm named SIFT-SVM is proposed to improve the recognition speed of high-resolution images.Firstly,high-resolution SAR images are divided into sub-images,and different categories of sub-images are selected as training samples,extract the SIFT feature of sample images to weaken the influence of azimuth change and speckle noise.Then,the class centers of SIFT features are extracted by k-means algorithm to construct visual dictionary,visualizes the visual word frequency histogram of sub-images to obtain the feature descripted by visual words.Finally,a classifier based on support vector machine(SVM)is designed to realize the recognition of different sea ice regions.The simulation results show that the proposed method has higher recognition rate and faster operation speed,the sea ice category label can be obtained efficiently.(3)To improve the accuracy of SAR sea ice image classification,a sea ice classification method based on region labeling MRF model is proposed.The image speckle noise is suppressed by Gamma MAP filter,then the sea ice statistic model of SAR image is established based on the MRF model,By using maximum a posteriori probability estimation theory,the problem of image classification is transformed into the optimization of label field and feature field joint probability function.We use the regional sea ice recognition result to set the initial label field and extract the image texture feature as the characteristic field.The weighting coefficient of the characteristic field is taken as a function of the annealing temperature,the influence of characteristic field on the classification result will adjust to the annealing temperature.Finally,the improved watershed algorithm is used to extract the edge information of the image,and combine the edge information with the final labeling result to get the sea ice distribution map.The simulation results show that the proposed method is more robust to noise and perform better in edge preserving,it can get more accurate classification result.