CryoSat-2雷达高度计海冰波形优选特征分类

CryoSat-2 radar altimeter sea ice waveform preferred feature classification

监测海冰的类型变化或厚度变化是较为有效的海冰监测方式,本文通过CryoSat-2雷达高度计对北极海冰类型开展研究。利用雷达高度计对北极海冰进行分类,一方面是难以挑选出最优特征参数,另一方面单一的雷达高度计数据难以实现海冰的相对精细化分类。针对以上问题,本文构建了一种利用卡方检验、互相关信息及Wrapper包装法组合的特征优选方法,利用该方法挑选出的特征子集(SSD+Sigma0+LTPP+PP+SK+LEW)结合随机森林分类法将北极地区的雷达高度计数据分为海水、一年薄冰、一年厚冰及多年冰。本文方法在训练集分类正确率为93.32%,验证集正确率为92.42%,Kappa系数为0.90,均优于其他特征组合,可以基本实现对北极地区海冰的有效分类,同时分类结果也有助于海冰厚度的反演。

Monitoring the type change or thickness change of sea ice is a more effective way to monitor sea ice.In this paper,the CryoSat-2 radar altimeter is used to study the types of Arctic sea ice.Using radar altimeter to classify Arctic sea ice,on the one hand,it is difficult to select the optimal characteristic parameters,on the other hand,it is difficult to achieve a relatively refined classification of sea ice with a single radar altimeter data.In view of the above problems,this paper constructed a feature selection method using the combination of chi-square test,mutual information and Wrapper packing method.The feature subset(SSD+Sigma0+LTPP+PP+SK+LEW)selected by this method is combined Random forest classification divides the radar altimeter data in the Arctic into seawater,one-year thin ice,one-year thick ice,and multi-year ice.The correct classification rate of this method is 93.32%in the training set,92.42%in the validation set,and the Kappa coefficient is 0.90,all of which are better than other feature combinations,which can basically achieve effective classification of sea ice in the Arctic region,and the classification results can also help in the inversion of sea ice thickness.