Geo-parcel-based Land Cover Classification Mapping And Uncertainty Analysis With Hyperspectral Remote Sensing Images

Hyperspectral remote sensing images can be used to carry out refining land cover mapping,and relevant technologies have been widely used in agricultural production,vegetation ecology and urban life.However,most of the existing methods are pixel-oriented or object-oriented analysis methods,which are lack of rationality and precision in mapping units,and limit the applied level of hyperspectral remote sensing land cover mapping.In addition,the reliability of remote sensing mapping is also very important,and the quantitative analysis of uncertainty in hyperspectral remote sensing mapping is of great significance to improve the application value of the results.In view of this,this thesis proposes the idea of geo-parcel level hyperspectral land cover classification mapping,and uses hyperspectral images of Majiwan Village in Xiongan New Area to carry out experimental verification and accuracy evaluation,and analyzes the advantages and disadvantages of mapping at different mapping unites by type accuracy and morphology accuracy,at the same time,information entropy method is used to calculate the uncertainty of geo-parcel level hyperspectral land cover classification mapping.The main research contents and achievements of this paper are as follows:(1)In order to evaluate the effect and accuracy of hyperspectral remote sensing land cover classification mapping,this paper carries out hyperspectral land cover classification mapping for pixels,objects and geo-parcels by using random forest,and then compares and analyzes the type accuracy and morphological accuracy of three different mapping units.In terms of type accuracy,the results show that hyperspectral remote sensing land cover classification at geo-parcel level has better type accuracy.In terms of morphological accuracy,the geo-parcel level mapping is significantly superior to the object.After the comprehensive comparison of the two types accuracy,it is concluded that the method of geo-parcel-oriented hyperspectral remote sensing land cover classification mapping can produce information results more in line with users’ needs,and has high potential application value.(2)Aiming at the problem of quantification and evaluation of uncertainty in geo-parcel level hyperspectral remote sensing classification mapping,this thesis further developed the calculation method based on information entropy,proposed three quantitative evaluation indexes of uncertainty,and verified the rationality of the indexes by using variance analysis and geo-detector,which can well indicate the reliability of geo-parcel land cover classification mapping at pixel level and geo-parcel level.Based on the above research,this thesis demonstrates the superiority of hyperspectral remote sensing land cover classification mapping at the geo-parcel level,which can form a mapping scheme more in line with users’ fine requirements.The main performance is to eliminate confusing information in pixel level mapping and improve mapping accuracy.In addition,the quantitative research on the uncertainty of land cover classification mapping makes up for the shortcomings of the traditional accuracy evaluation,and realizes the reliability analysis of the mapping in the whole domain,which provides an important reference for the quality evaluation and optimization of the mapping.