Research On Extraction Of Thermokarst-Lake In Qinghai-Tibet Plateau Based On Sentinel Remote Sensing Image

Thermokarst-lake are typical landforms in permafrost regions.In recent years,due to the impact of human activities and global warming on the ecology of the Qinghai-Tibet Plateau,Thermokarst have gradually occurred in permafrost regions,and a large number of small lakes have been generated in the permafrost regions.It is called thermal karst lake.Thermokarst-lake is a direct channel for permafrost to emit greenhouse gases and is closely related to the regional groundwater level,which feeds back to climate change and afects the evolution of the regional ecological environment.The plateau climate in the Qinghai-Tibet Plateau is warming year by year.Human engineering Activities gradually expanded in the plateau area,which had a great impact on the fragile ecological environment of the Qinghai-Tibet Plateau.The Qinghai-Tibet Plateau is a mapping of the global climate and has a mutual feedback effect with the world's climate environment.Therefore,it is of great scientific significance to carry out related research on thermal melting lakes.In recent years,with the continuous development of remote sensing technology,land surface water monitoring based on satellite images has gradually replaced traditional manual monitoring.From the single-band threshold method to the multi-band water index method,water monitoring has undergone a transition from manual to fully automatic extraction.Due to the large differences in the spectral characteristics of the Thermokarst-lake in remote sensing images and the small area of Thermokarst-lake,and there are features similar to the hot melt lake spectra in the hot melt lake development area,there is still a lack of effective large-scale extraction of heat.Therefore,this research is based on sentinel remote sensing images to study the extraction algorithm of hot melt lake.The article optimizes the inter-spectral relationship method in the multi-band method in the traditional method,and proposes the inter-spectral relationship based on LBV transformation.Relational model;and introduced the Rotation Forest algorithm(Rotation Forest,RoF)and deep learning DeepLab v3+model for the study of hot melt lake extraction.In this study,the following conclusions were obtained by studying the extraction algorithm of the thermal melting lake information in the image:1.Study the Coastal aerosol band,Green band,NIR band,SWIR1 band using Sentinel-2A image data,and its wavelength value ?1=0.443?m,?2=0.56?m,?3=0.842?m,?4=1.610?m.According to the characteristics of the Thermokarst-lake,according to the spectral brightness value of the ground object samples,combined with the B formula in the LBV transformation formula,a linear regression equation is established,which is used as a model for the relationship between the spectra to extract the hot melt lake information.2.The study introduces the rotating forest learning algorithm,referring to the characteristics of the hot melt lake,and according to the influence of feature rotation,the more commonly used unsupervised principal component analysis method is used as the feature extraction method,which is used as the spectral feature of the Thermokarst-lake to conduct thermal Extraction of melting lake.3.This study uses the DeepLab v3+model of deep learning because it can maintain performance while greatly reducing the amount of calculation.DeepLab v3+adds the Decoder module to improve edge information to capture more scale information,such as the boundary information of hot melt lakes,and further improve the accuracy of Thermokarst-lake extraction.Its extraction accuracy is far superior to the rotating forest algorithm and the LBV spectrum relationship method.Figure[29]Table[9]Reference[80]...