Urban Boundary Defined On The Basis Of Sentinel 2 And POI Data

It is the premise and foundation to delimit the city boundary reasonably for controlling the orderly growth of a city.Accurate demarcation of urban boundaries helps us to understand the overall development law of urban space to understand the spatial differentiation law of urban internal and external development.It also can guide the rational expansion of cities.It is of great significance to the high-quality development of urbanization in China.As the capital city of Henan province,Zhengzhou has a rapid urbanization process and an increasingly intensified urban expansion.Therefore,it is urgent to rationally plan the development of Zhengzhou,optimize the urban spatial layout and improve the quality of urbanization.It took Zhengzhou city as the research area and collected socio-economic data for 2018 and Sentinel-2,high-resolution multispectral remote sensing image data,for 2018.It used the random forest model,to extract of Zhengzhou urban construction land based on ENVI 5.3 software and Arc GIS platform for data processing and analysis and drew the boundary of Zhengzhou city in 2018.The main research contents are as follows:(1)Sentinel 2 and POI data feature extraction.We downloaded Sentinel-2 high resolution multispectral images of Zhengzhou in 2018 with less than 5% cloud cover from the USGS.Due to the large scope of the research area in Zhengzhou,one image cannot cover the whole area.In this paper,eight high resolution multispectral images are selected.In addition,due to the influence of image cloud size,some months may not have suitable images for research and analysis,so we chose images of nearby months to replace them and try to shorten the time between two images to be synthesized into one image as much as possible.We produce four remote sensing images of Zhengzhou in February,June,September and November 2018 and unify the spatial resolution of the image to 10 m.We analyze the influence of band 2,band 3,band 4,band 5,band 6,band 7,band 8 and band 8A of high-resolution multi-spectral remote sensing image on urban boundary.It extract texture features(including mean,variance,contrast,homogeneity,dissimilarity,correlation,entropy and second moment),normalized difference vegetation index(NDVI)and normalized difference building-up index(NDBI)of Zhengzhou in 2018 by using Sentinel-2 images.The real-time POI big data is obtained through Baidu map.After the interest points of the same location attribute are eliminated,they are classified according to their attributes and divided into 5 categories,including administrative center,residential district,public service,commercial center and transportation.The characteristics of POI data,including distance to administrative center,distance to residential district,distance to public service,distance to commercial center,distance to transportation,were extracted by using Kernel Density and Euclidean Distance analysis method.A total of 263 features were selected as input variables for model training.(2)Extraction of urban construction land.It based on random forest(RF)algorithm to establish the extraction model of urban construction land.By setting up multiple sets of training data with different quantities,the extraction model of urban construction land was trained respectively,and it obtained different training results.The results of each training set were verified by establishing the confusion matrix and calculating the Kappa coefficient.In the 13 rd group,there were 130 training samples participating in the model training,and the model with 70 verification samples.It showed the best training effect and the highest overall accuracy of the model,which is 96.25%,and the Kappa coefficient was 0.93.Compared with Support Vector Machine(SVM)algorithm,the experimental results showed that the results of the urban construction land extraction model training based on random forest algorithm were more consistent and more accurate.(3)Urban boundaries demarcation.As the raster image of urban construction land extracted by the model tends to be in one or more closed areas.There are some cavities inside the city,and there are a lot of noises outside the main patches of urban construction land.The boundary area is relatively broken with relatively broken patches.In order to obtain a complete and smooth boundary of urban construction land,the extraction results of urban construction land were optimized based on mathematical morphology method.After several close and open operations,the structural element of 13×13 size was finally selected to normalize the boundary.And it was extracted a smooth,continuous and more realistic urban boundary.By comparing with the FROM-GLC10(Zhengzhou)and the 2018 visual interpretation results,the overlapping grid numbers of the matching area and overlapping ratio were calculated.We analyzed and the range and location accuracy of the demarcation results.On the whole,the matching value between the delineation result and FROMGLC10(Zhengzhou)was 117.57%,and the matching value between the delineation result and the visual interpretation was 122.13%.The overlap ratio of position is 89.98% and 90.66% respectively.The delineation boundaries has high consistency in both scope and location.