Study On Multi-source Land Cover Data Fusion Method Based On Superpixel

Accurate large-scale land cover map can provide basic data support for scientific research on the relationship between natural and biological activities,land cover and spatial pattern,simulation monitoring and evaluation of ecological environment change,and human social and economic development.It is a challenging task to draw large scale land cover map with high precision and high resolution by using remote sensing images.It is considered an effective method to integrate the information of existing products while acquiring new products.Fusion process,however,tend to retain the different product information at lower spatial resolution scales,lead to errors in information,at the same time take from low spatial resolution land cover data sample will decrease because of the mixed pixels image sample accuracy Therefore need to study a kind of can solve the land cover data fusion method,a new method of limit.To solve this problem,we propose a multi-source land cover product that combines8 multi-source land cover products,CCI-LC,CGLS,FROC-GLC,MCD12Q1,GFSAD30,PLASAR,GSWD and GHS-BUILT,and propose a superpixel-based product Principal component analysis(PCA)and statistical extraction technology of multi-source fusion product mapping method combined with the Google earth engine(GEE)huge amounts of data and machine learning algorithms,purification of coarse consistency error information in the area,obtain reliable training samples Accordingly,inconsistency of multi-source products area correction,eventually merge the best resolution of existing products(30 M)After a series of studies,the following results are obtained:(1)Based on the superposition analysis of spatial consistency,it is found that the pixel proportion of the highest consistency level in the study area is only 45.42%.The area consistency between each product is low and the correlation between land cover products is weak.With the increase of scale,large amounts of information loss in land cover products will inevitably affect the collaborative application Analysis based on the inconsistency,they found that feature category is the degree of consistency as a product of the classification error on the low side,and cause systemic factors,which affect the consistency include the establishment of the remote sensing image data classification system and classification method.(2)According to the characteristics of low consistency of multi-source land cover data and the factors that affect the inconsistency,a multi-source land cover data fusion method based on superpixel is proposed.This method is an automatic fusion strategy,and seamless time series images of the study area are synthesized based on GEE platform and feature selection is established.Consistency was proposed based on pixels of rough area after excluding method,the method in constructing the super level pixel scale model of top-down segmentation,the consistency of a wide range of rough area and through principal component analysis of rough image for dimension reduction,the consistency of the area according to the statistics excluded from group deserves to have fine consistency area(30 m);Finally,local adaptation samples were used to correct the inconsistent areas,and the overall accuracy of the new land cover product fusion was85.80%.Compared with the existing CCI-LC,CGLS,FROM-GLC and MCD12Q1multi-category land cover products,the accuracy of the fusion result was improved by11.75-24.17% Kappa coefficient was 0.82,0.16 higher than 0.3 with GFSAD30,PLASAR,GWSD and GHS-BUILT The overall accuracy of fusion results was improved by 2.99-20.71% and the Kappa coefficient was improved by 0.22-0.56 compared with the four single category products,and the over-interpretation of single category products in inconsistent regions was corrected.Based on the existing multi-source land cover products,the method proposed in this study can quickly integrate the effective information of multi-source land cover products and provide an effective method for obtaining high-precision and high-spatial resolution land cover products At the same time,with the open access of remote sensing data and the increase of different types of land cover products,the analysis and fusion method of multi-source land cover data proposed in this paper will be more useful for reference.