冰川识别对于周边地区水资源与气候变化监测具有重要意义。全极化SAR影像包含丰富的特征，而深度学习能够充分挖掘影像信息，因此使用全极化SAR影像结合深度学习能够得到良好的冰川识别效果。本文基于喜马拉雅山脉西端ALOS2-PALSAR全极化影像，使用VGG16特征提取网络与全卷积神经网络模型U-net相结合的VGG16-unet 对冰川进行识别。采用的特征包括极化相干矩阵对角线元素、Freeman-Durden、H/A/α、Pauli、VanZyl、Yamaguchi五种极化分解参数共计19种特征。为了充分利用影像信息，对这些特征进行分析与组合，并比较它们之间的冰川识别精度，以选取最佳特征。由于冰川与非冰川的地形具有明显差异，因此将DEM、坡度、局部入射角等作为辅助特征与极化特征结合。通过对比不同极化特征分类精度得出，基于物理特性的Pauli、Freeman-Durden、VanZyl、Yamaguchi特征分类的精度较高，其中Pauli特征分类的精度最高，整体精度(OA)达到92.54%，平均用户交并比(mIoU)达到78.78%。加入地形数据后整体精度(OA)提升至94.34%，平均用户交并比(mIoU)提升至82.35%。为了进一步提高冰川的识别精度，提出了一种基于单波段特征整体精度(OA)及召回率(Recall)筛选出的SDV特征交叉组合方式，结果显示，该组合整体精度(OA)达到94.98%，用户交并比(mIoU)达到85.67%，比Pauli特征分类精度分别高出0.64%和 3.32%。上述结果表明，选择最佳的特征组合方式并结合深度学习在提升冰川识别精度中具有重要的作用。

Glacier identification is important for monitoring water resources and climate change in surrounding areas. Fully polarized SAR images contain rich features, and deep learning can fully exploit the image information, so using fully polarized SAR images combined with deep learning can get accurate glacier recognition results. In this paper, we use VGG16-unet(VGG16 combine with U-net)to identify glaciers based on ALOS2-PALSAR fully polarized images of the western part of the Himalayas. The features include the diagonal elements of the polarization coherence matrix, Freeman-Durden, H/A/α, Pauli, VanZyl, and Yamaguchi polarization decomposition parameters totaling 19 features. In order to make full use of the image information, these features are analyzed and combined, and the glacier recognition accuracy is compared between them to select the best features. Since there are obvious differences between glacier and non-glacier topography, elevation(DEM), slope, and local incidence angle are combined with polarization features as auxiliary features. By comparing the classification accuracy of different polarization features, it is concluded that the accuracy of Pauli, Freeman-Durden, VanZyl, and Yamaguchi features based on physical characteristics is higher, among which Pauli features has the highest accuracy with an overall accuracy (OA) of 92.54% and an average user intersection ratio (mIoU) of 78.78%. The overall accuracy (OA) was improved to 94.34% and the average user intersection ratio (mIoU) was improved to 82.35% after adding the topographic data. To further improve the recognition accuracy of glaciers, a feature cross-combination approach is proposed, and the results show that the overall accuracy (OA) of the combination reaches 94.98% and the user intersection ratio (mIoU) reaches 85.67%, which are 0.64% and 3.32% higher than the classification accuracy of Pauli features, respectively. The above results show that choosing the best feature combination method and combining with deep learning plays an important role in improving the accuracy of glacier recognition.