中文摘要：

以新疆玛纳斯河流域2014年3月19日RADASAT-2影像为研究数据,采用马尔可夫随机场（Markov Random Field,MRF）模型分割方法进行积雪识别.MRF模型分割方法能够充分利用图像上下文信息,降低相干斑噪声对合成孔径雷达（Synthetic Aperture Radar,SAR）数据的影响.通过初始k-means分割估算出MRF参数,建立先验模型和概率密度函数,利用迭代条件模式（Iterated Conditional Model,ICM）算法进行最大后验概率求解得到最优标记,从而识别出积雪.通过实测数据进行验证,该方法积雪识别精度达86.67%.结果表明：MRF模型分割方法的能够有效识别积雪;在地势较为平坦的地区,交叉极化（HV）方式下的后向散射系数与极化总功率Span的识别效果较好;在地形起伏较大的地区,HV后向散射系数的识别效果随着高程和坡度的增加而降低,极化总功率Span能够综合三种极化特征,较好地克服地形影响,提高积雪的识别精度.

英文摘要：

This study proposed Markov Random Field（MRF）to recognize snow cover using RADARSAT-2data on 19 March 2014in Manasi River Basin,Xinjiang Province.The MRF model based image segmentation method can take full advantage of the contextual information,and reduce the influence of speckle noise on SAR data.We estimated the MRF parameters following the initial k-means segmentation,then established the prior model and the probability density function.Finally,we used Iterated Conditional Model（ICM）for solving the maximum posterior probability of the optimal label to identify the snow cover.Verified by the field survey data,the accuracy of the method to recognizesnow cover was 86.67%.The results showed shat the MRF model based segmentation method could effectively recognize snow cover.In the flat areas,the backscattering coefficient under the cross-polarization HV and the polarization total power Span had the good recognition accuracy.But in the mountainous areas,the recognition accuracy of the HV backscattering coefficient decreased with the increase of elevation and slope.The polarization total power Span can integrate the three polarization characteristics to overcome the topographic effect,and to impove the recognition accuracy of snow cover.