Reply on RC1

Abstract

Accurate and quick extraction of glacier boundaries plays an important role in studies of glacier inventory, glacier change and glacier movement, and it faces great opportunities and challenges due to the increasing availability of high-resolution remote sensing images with larger data volume and richer texture informations. In this study, we improved the DeepLab V3+ as Attention DeepLab V3+ and designed a complete solution based on the improved network to automatically extract glacier outlines from the Gaofen-6 PMS images with a spatial resolution of 2 m. In the solution, the Test-Time Augmentation (TTA) was adopted to increase model robustness, and the Convolutional Block Attention Module (CBAM) was added into the Atrous Spatial Pyramid Poolin (ASPP) structure in DeepLab V3+ to enhance the weight of the target pixels and reduce the impact of useless features. The results show that the improved model effectively improves the robustness of the model, enhances the weight of target image elements and reduces the influence of non-target elements. Compared with deep learning models such as FCN, U-Net and DeepLab3+, the improved model performs better, with OA and Kappa coefficients of 99.58 % and 0.9915 for the test dataset, respectively. Moreover, our method achieves the highest OA and Kappa of 99.40 % and 0.9846 for glacier boundary extraction in parts of the Tanggula Mountains and Kunlun Mountains based on Gaofen-6 PMS images, showing its excellent performance and great potential.