利用分层特征组合策略的全极化SAR山区积雪识别

Snow Cover Recognition in Mountainous Area by Full-polarimetric SAR Using Layered Feature Combination Strategy

积雪作为冰冻圈重要组成部分,与全球气候变化和生态系统密切相关,精准识别积雪分布信息具有重要意义。合成孔径雷达(Synthetic Aperture Radar,SAR)数据的极化和散射特征在积雪识别中具有极大的应用潜力。以新疆玛纳斯河流域为研究区,提取全极化Radarsat-2数据后向散射特征和目标极化分解特征;为探索极化特征和散射特征对积雪识别的贡献,将获取的特征进行组合,得到3种特征集;采用随机森林算法对研究区积雪进行识别提取。结果显示,基于随机森林的Radarsat-2极化特征结合散射特征识别结果的总体精度和调和平均值(F1)达到最高,分别为83.00%和0.82,仅基于极化特征识别结果总体精度和F1分别为77.5%和0.76。研究结果表明,与单一极化特征相比,结合散射特征和极化特征能有效提高积雪识别精度,对山区大范围积雪识别具有极大的潜力。

As an important part of cryosphere,snow is closely related to global climate change and ecosystem.Accurate recognition of snow distribution information is of great significance.The polarization and scattering features of Synthetic Aperture Radar(SAR)data have great application potential in snow cover recognition.Taking Manas River Basin in Xinjiang as the study area,the backscattering features and target polarization decomposition features of fully polarized Radarsat-2 data are extracted.In order to explore the contribution of polarization and scattering features to snow cover recognition,the obtained features are combined to obtain three feature sets.Finally,the random forest algorithm is used to identify and extract the snow cover in the study area.The results show that the overall accuracy and harmonic mean(F1)of Radarsat-2 polarization feature combined with scattering feature recognition based on random forest reach the highest values,which are 83.00%and 0.82 respectively,and the overall accuracy and F1 based on only polarization feature recognition are 77.5%and 0.76 respectively.The research results show that compared with single polarization feature,the combination of scattering feature and polarization feature can effectively improve the accuracy of snow cover recognition,which has great potential for large-scale snow cover recognition in mountainous areas.