Hierarchical-based Land Cover Classification Based On Multi-source Remotely Sensed Data

The types of urban land cover affect the thermal environment and human settlement environment.Mapping the distribution of urban land cover types quickly and accurately is of great significance to the study of urban heat island effect,the evaluation of the service value of urban ecosystem and the formulation of sustainable urban development policies for relevant departments.The detailed classification of urban land cover is based on remote sensing data with high spatial resolution,and expert knowledge rules and machine learning algorithms are used for classification.Due to the complexity of land cover types and the limitation of high-resolution remote sensing data in spectral bands,it is difficult to effectively distinguish different forest types and buildings from other impervious types.In this study,Jinjiang City and Shishi City in Fujian Province were taken as the study areas to explore the fine classification of urban surface based on multi-source high-resolution satellite data.The multispectral band and panchromatic band of ZYU-3were fusion to extract fine spectral information.Taking the stereo data of ZYU-3 as the data source,the urban surface canopy height model(CHM)was produced by combining the control points selected from the ground in the research area and the inverse distance weight interpolation method.On this basis,the height of urban features was extracted.A data scheme containing only spectral information(ZY3)and a data scheme combining spectral and ground object height information(ZY3+CHM)were designed,and a hierarchical classification method(HBC)based on the hierarchical system and cyclic optimization variable set was proposed,which was compared with the random forest classifier.The main results are as follows:(1)The classification scheme combined with the ZY3+CHM data scheme and the hierarchical classification method was the optimal data scheme.The study area was divided into 11 categories,including eucalyptus,casuarina,other broad-leaved,Spartina interflora,grassland,crops,buildings,non-building impervious,bare soil,tidal flats and water bodies.The overall classification accuracy of vegetation part was 81.92%.The Kappa coefficient was 0.76,and the overall classification accuracy of non-vegetation part was 89.55%,and the Kappa coefficient was 0.86.(2)CHM can effectively improve the classification accuracy of urban land cover.In the urban vegetation area,CHM improved the classification performance of eucalyptus,other broad-leaved trees and Spartina alterniflora most significantly.When the hierarchical classification method was applied,the user accuracy of other broad-leaved trees increased by 14.73%,and the producer accuracy increased by 15.63%.In urban non-vegetated areas,CHM improved the classification performance of buildings and other impervious areas most obviously.When the hierarchical classification method was applied,the user accuracy of buildings increased by 11.45% and producer accuracy increased by 36.63%.(3)In the fine classification of urban land cover,the hierarchical classification method can not only intuitively feed back to users to distinguish the key variables of each land cover type,but also obtain better classification performance in different variable schemes,and has certain advantages compared with random forest method.In the urban vegetated area,the overall classification accuracy is up to 3.08%,and the Kappa coefficient is up to 0.04.In the urban non-vegetated area,the overall classification accuracy is up to 2.26%,and the Kappa coefficient is up to 2.26%.