Remote Sensing Classification Method Of Swamp Based On Sentinel-1/2 And Multiple Machine Learning Algorithms

Wetlands are valuable ecosystems on Earth,with a variety of unique functions and services.Not only do they provide humans with a large amount of raw materials and water resources,they also maintain ecological balance,protect biodiversity and rare species resources.As an important part of wetlands,swamp have extremely strong carbon sequestration function and rich carbon storage,which is of great significance to the global carbon and oxygen balance.At the same time,wetlands are also an important component of the cold region of northeast China.Due to the degradation of permafrost for many years in the cold region,forest and wetland ecosystems are bound to be affected,and the conversion between bog and woody vegetation is intensified,so a fine classification of swamp wetlands can provide an effective reference for monitoring and management of wetlands.With the development of remote sensing technology,the classification of swamp information using remote sensing data has been widely used because the wetland environment is complex and some places are difficult to reach by human.In this paper,we take the Emur River basin in Daxinganling,which is rich in swamp resources,as the study area,and combine Sentinel-1 and Sentinel-2multi-source remote sensing data to establish various feature combination plans of spectral features,red-edge features and radar features,and use Support Vector Machine（SVM）,Random The results and accuracies of the nine classification plans were used to explore the effects of different feature variables on the classification effect;meanwhile,multiple machine learning methods were compared to select the optimal classification method for swamp classification.The main findings of this paper are as follows.（1）Sentinel-1/2 multi-source data establish multiple feature combination plans for wetland classification accuracy is significantly improved,with the addition of red-edge features and radar features,the accuracy of marsh wetland classification is gradually improved,and the red-edge band and radar VV and VH polarized backscatter coefficients are very helpful for the identification of swamp and herbaceous marshes.The study area was classified into nine types of land types:swamp,herbaceous swamp,water,residential land,woodland,cropland,grassland,dryland,and road,and classified using three feature plans,plan 1 using only spectral features containing Sentinel-2 bands and vegetation indices,plan 2 adding seven red-edge indices,and plan 3 adding two more radar backscatter coefficients of VV and VH.Comparing the classification accuracies of SVM,RF,and DL methods,all of them are that the classification accuracy of plan 1