A Glacier Identification Method Based on Fused Multimodal Remote Sensing Images
Glaciers are extremely sensitive to climate change,and changes in glaciers are closely related to regional ecology,natural disasters and water resources.Real-time monitoring and information extrac-tion from remote sensing of highland glaciers is an indispensable means to monitor glacier changes.In or-der to effectively identify glaciers in multi-scale high-resolution remote sensing images,a Glacier-Unet model is designed.The specific work is(1)to address the existing Landsat satellite remote sensing image based plateau glacier extraction algorithms due to the lack of response to the impact of complex feature in-terference,resulting in the loss of reflective target information.Taking the Tibetan Plateau Animaqing Snow Mountain as a test object,a dataset based on Landsat-9 remote sensing satellite high-resolution im-age production was selected.Data preprocessing is carried out on high-resolution glacier remote sensing images,feature-level fusion and pixel-level fusion are adopted to produce multimodal remote sensing data images,and semantic segmentation datasets are enriched by sliding slices and data enhancement means to ensure the accuracy and robustness of the model training;(2)Aiming at the insufficient recognition abil-ity of scattered and tiny glaciers,a gated multi-scale filter layer(G-MsFL)is designed.filter layer(G-MsFL)is designed to filter out useless feature information.The model is equipped with multi-scale fea-ture extraction and feature fusion capabilities,which can effectively identify glaciers in complex feature environments;(3)Paralleling Dual Attention Module(P-DAM)is designed to address the problem of fuzzy glacier contours.The rich contextual information of the glacier boundary is encoded as the local fea-tures of the feature map,thus enhancing its feature expression ability.Qualitative and quantitative analy-ses of the experimental results of the improved Glacier-Unet model in the test dataset reveal that the over-all segmentation accuracy is improved by 6.1％compared with the comparison method,and it can effec-tively identify scattered and tiny glaciers,which is of great significance for the glacier identification work in the plateau region.
NETL
NSTL
万方数据
