Research And Application Of Lake Ice Classification Based On HMRF-EM Method

Lake ice,as a component of the cryosphere,is one of the key water cycle parameters of the Earth's surface.Its change in time series can relatively intuitively reflect the climate change in the area where the lake ice is located.Therefore,the study of lake ice is of great significance in the study of regional climate change.At present,research on lake ice classification at home and abroad can be divided into lake ice classification based on active remote sensing data and lake ice classification based on passive remote sensing data.Lake ice classification based on active remote sensing data can classify different types of lake ice.The hidden Markov random field and expected maximum(HMRF-EM)algorithm is a classification method that can effectively classify SAR images.However,there is no relevant research on lake ice classification method based on this algorithm.Therefore,this study proposed a lake ice classification method based on HMRF-EM algorithm which is relying on the Earth Big Data Science Engineering Spatio-Temporal Tripolar Environment Project,and classified part of the lake ice in the Lena River Delta from January to April 2018.Due to the limitation of geographical conditions and other factors,there is currently a lack of field observation data for the monitoring of lake ice changes in the study area.Therefore,in this study,simulated data was used instead of real data for parameter selection and accuracy evaluation of HMRF-EM algorithm.The parameter selection includes the selection of optimal parameters for lake ice classification,and verifying whether different initialization methods have an impact on the accuracy of classification results.The result of parameter selection shows that when the potential energy parameter ? is in the range of [0.1,2],the classification result of the HMRF-EM algorithm using the K-Means algorithm and the fuzzy C-means clustering algorithm as the initialization method is close to the real classification result,when the lake ice classification effect is the best,the initialization method of the HMRF-EM algorithm is the K-Means algorithm,and the potential energy parameter ? is 0.1,its overall accuracy is 99.76%,Kappa coefficient is 0.995.And different initialization methods in the lake ice classification will have a certain impact on the classification accuracy of the HMRFEM algorithm.In the classification accuracy evaluation,the threshold method,parallelepiped method,K-Means method and Iterative Selforganizing Data Analysis(ISODATA)method was selected to compare with the HMRF-EM algorithm.Compared with other methods,the overall classification accuracy of the HMRF-EM algorithm is improved by 1%-23%,the Kappa coefficient is improved by 0.03-0.53,which means the HMRFEM algorithm is better than other algorithm in the lake ice classification.In the end,the Sentinel-1 satellite data was selected as the main data source to classify lake ice,divided into two types of floating ice and grounded ice,and satisfied the changes of different types of lake ice in time series.It can be seen from the results that as time changes,the area of floating ice gradually decreased,and the area of grounded ice gradually increased,which is consistent with the winter temperature conditions in the Siberian polar region,that proves the HMRF-EM algorithm can classify the different types of the lake ice effectively.