Research On The Extraction Method Of Impervious Surface Information From UAV Low-altitude Remote Sensing Images

The study of impervious surface was of great significance for urban planning,urban ecological civilization construction,and construction of sponge cities.With the advancement of aerial remote sensing technology,traditional satellite remote sensing technology has gradually shifted to miniaturized and lightweight unmanned aerial vehicle(UAV)remote sensing technology.UAV remote sensing can obtain highresolution images at low cost and efficiently.At the same time,UAV low-altitude image matching point cloud,as a new type of geospatial data,occupy an increasingly important position in the construction of smart city spatial data framework.However,the difficulty in accurately and efficiently extracting the impervious surface from the orthophotos obtained by the UAV remote sensing technology is that the UAV low-altitude remote sensing image has relatively fewer bands than the satellite remote sensing image,and the available spectral range exists only in visible-bands.In this range,it is more difficult to extract the impervious surface.Therefore,it is imperative to study a method that can quickly extract impervious surfaces from UAV remote sensing images.This study analyzed the spectral feature space of different waveband combinations and constructed a visible-bands remote sensing index.Based on this,combined with the matching point cloud,a new method is proposed to apply to the UAV remote sensing impervious surface information extraction.The main research work of this paper is as follows:(1)Using the Landsat8 OLI image as the data source,the impervious surface is extracted by using six indexes: IBI,NBI,NDBI,PII,RRI,and UI.Based on the analysis of the extraction results and accuracy,this study used the same linear combination index construction method as the PII index construction method to establish a Green-Blue spectral feature space in the visible-bands.The combination of soil line and impervious surface line can construct the Green-Blue impervious index(GBISI)for the effective separation of vegetation pixels and impervious surface pixels.In order to verify the accuracy of the GBISI on the impervious surface,the satellite image of GF-2 was used as the data source,and the PII and RRI with better extraction results from the six indexes were used to conduct the impervious surface respectively extracted.The results show that the overall accuracy of the GBISI extraction result reached 95%,which was higher than the overall accuracy of the RRI extraction.The validity of the GBISI constructed in the visible-bands was verified,which provided support for the extraction of impervious surfaces from UAV remote sensing images.(2)Based on structure from motion(SFM)and CMVS/PMVS dense matching algorithm,reconstruct dense point clouds of low-altitude images in the study area and generate digital surface models(DSM),and then use cloth simulation filtering(CSF)method to obtain the ground points and non-ground points,a digital elevation models(DEM)is generated by interpolating the ground points,and the normalized digital surface models(n DSM)is represented by the difference between DSM and DEM.The n DSM threshold and the GBISI in the study area were determined according to the n DSM value statistical graph and scatter plots of different features in the Green-Blue spectral feature space.Finally,the n DSM obtained from the matched point clouds combined with the GBISI was used to extract the impervious surface of the loworthophoto image of the UAV,the results were compared and analyzed with the results obtained by simply using the VDVI method and the GBISI method.The accuracy evaluation results show that GBISI combined with n DSM method has the highest overall accuracy for impervious surface extraction,reaching 96.95%,which is 2.85% higher than the extraction accuracy of GBISI method,and 11.33% higher than the extraction accuracy of VDVI index method.At last,this study used n DSM to subdivide the impervious surface into buildings and concrete pavements,and mapped the impervious surface.The research results in this study show that the GBISI constructed by the GreenBlue spectral feature space can effectively separate soil pixels and improve the accuracy of impervious surface extraction.Benefit by the low-altitude photography of UAV,which can obtain the characteristics of terrain features,the introduction of matching point clouds not only solved the problem of misclassification caused by vegetation cover on the roof of the building,but also further subdivided the impervious surface into buildings and road surface.The research proved that the method of extracting the impervious surface based on the matching point clouds and GBISI can be used as a new method to extract the impervious surface information of the UAV remote sensing image.