Research On Super-resolution Snow Identification On The Qinghai-Tibet Plateau Based On Deep Learnin

The Qinghai-Tibet Plateau is one of the areas with the most snow accumulation in China,and is known as the snow-covered plateau.The change of snow accumulation in the QinghaiTibet Plateau is inseparable from water resource management,climate change and the production and life of people in the plateau and downstream areas.Due to the harsh climate and sparsely populated areas on the plateau,it is difficult to obtain the temporal and spatial distribution of snow in these areas.With the development of satellite remote sensing technology,more and more optical satellite data are used for snow area monitoring,but most satellite sensors cannot meet the requirements of large-scale and real-time snow area monitoring due to spectral and spatial resolution constraints.China’s new-generation stationary orbit satellite FY-4A has the characteristics of fixed observation points and high temporal resolution,which is conducive to real-time,large-scale and accurate monitoring of snow area.However,the spatial resolution of stationary orbit satellites is usually much lower than that of geostationary satellites,which makes it difficult to obtain high spatial resolution snow products,thus severely limiting the application range of FY-4A satellite images.This paper takes FY-4A/AGRI remote sensing data as the data source and uses deep learning methods to enhance the spatial resolution of the remote sensing data.Then,based on the high spatial resolution remote sensing data and combined with geospatial information data and the high temporal resolution characteristics of FY-4A to discriminate the snow of the Qinghai-Tibet Plateau.The main work is as follows:(1)Remote sensing image super-resolution reconstruction method based on matching extraction and cross-scale feature fusion network.In view of the fact that the spatial resolution of the near-infrared and shortwave infrared bands of FY-4A/AGRI data is much lower than that of the corresponding visible light bands,resulting in the problem of low spatial resolution of the generated snow products,this paper proposes a remote sensing image super-resolution network based on feature matching extraction and cross-scale feature fusion.The network takes high-resolution band images as reference images,uses the matching extraction module to match according to the local similarity between high-resolution images and low-resolution images,and then uses the cross-scale feature fusion module to fuse the extracted high-resolution image features with low-resolution image features,combined with the spatial-spectral total variation loss to ensure the spatial and spectral credibility of the reconstruction results.The experimental results show that the proposed method has good performance in both visual effect and objective evaluation indicators,providing fine and accurate spatial information while preserving the spectral information of the original bands,which can expand the application scope of FY-4A data.(2)Cloud and snow recognition method based on edge enhancement and attention mechanism.In order to solve the problem of low recognition accuracy of existing snow products,fuzzy boundaries and easy misjudgment of cloud-covered pixels as snow pixels,FY-4A/AGRI remote sensing data after super-resolution reconstruction are used as input,and edge enhancement branch and edge loss function are used to improve the recognition effect of cloud and snow edge areas and fragmented cloud and snow.In view of the characteristics of many input channels,different importance of different channels for cloud and snow segmentation,and small difference between clouds and snow,the enhanced classes coordinate attention module is used to improve the prediction accuracy of clouds and snow by combining the relationship between channels and long-distance location information.Finally,multi-temporal cloud filtering is used to make full use of the high temporal resolution characteristics of FY-4A/AGRI and reduce cloud coverage to restore the snow surface covered by clouds.Comparing with mainstream snow products and verifying with ground meteorological station data,it is demonstrated that snow identification on remote sensing data after super-resolution reconstruction has achieved good recognition effect,which greatly improves the spatial resolution of snow products and effectively improves the misjudgment of cloud and snow.