Extracting Land Cover Types From Orthophotos And LiDAR Data

With the development of remote sensing technology,manual,semi-manual and automated ground feature extraction based on remote sensing images has replaced manual field surveys as the most important land cover information extraction method.Due to the problems of ground object classification using a single data source,multi-source remote sensing data have been widely used for extracting and updating land cover information.The eastern part of Surrey in the Greater Vancouver area of British Columbia,Canada was used as the study area.This study combined the spectral,shape,texture and height features of orthophotos and LiDAR data in 2009 and 2013 for object-based land cover classification,and analyzed land cover changes from 2009 to 2013.The classification process includes four major steps:(1)Construct a multi-scale segmentation system.The red,green,and blue bands of orthophotos and the normalized digital surface model(n DSM)and normalized difference image(ND)derived from LiDAR data were used as the input data for multi-scale segmentation.(2)Determine the classification rules.The spectral,shape,texture and height information of the input data were used to determine the membership classification rules.(3)Determine the optimal classification method through comparative experiments.The first method is to perform object-based membership classification and SVM classification to extract land cover information by combining the spectrum,shape,texture,and height characteristics of orthophotos and LiDAR data,the average classification accuracy both reaches 92%;the second method uses the spectral,shape,and texture features of orthophotos to perform object-based membership classification and SVM classification to extract land cover information.The average classification accuracy reaches 82% and 80%;the third method is to perform object-based membership classification and SVM classification to extract land cover information by combining the spectrum,shape and height features of orthophotos and LiDAR data.The average classification accuracy reaches 87% and 83%.Through the comparison of the above classification methods,it is determined that an object-based membership classification method combining the spectrum,shape,texture and height characteristics of orthophotos and LiDAR data is adopted,and forest,grass,building,road,parking-lot are successfully extracted.(4)Analysis of land cover changes.There were substantial changes in land cover from 2009 to 2013,and the forest area was reduced and transformed into buildings.The research results show that the object-based classification method using orthophotos and LiDAR data can effectively reduce the limitations of using orthophoto spectral information for land cover classification,and reduce the occurrence of "different objects with the same spectrum" and "the same thing but different spectrum".The expansion of urbanization in the study area from 2009 to 2013 led to the replacement of urban woodland by buildings.The classification method based on the combination of orthophoto and LiDAR explored in this research provides a reference for relevant departments to accurately extract land cover information and other ground feature information,as well as for other scholars to carry out similar research.The results of land cover change analysis should,to some extent,provide scientific basis for land planning and environmental protection.