Extraction And Analysis Of New Built-up Land In Nanjing Based On Multi-temporal Remote Sensing Imagery

New built-up land is an important part of urban construction,and its accurate extraction can help urban planners and managers master the trend and development speed of urban development and realize dynamic monitoring and systematical analysis.The ESA Sentinel series of remote sensing satellites,which share 10m resolution visible,near-infrared and backscatter data free of charge for society,can be used to extract new built-up land at low cost,accurately and efficiently.In order to extract new built-up land accurately,an object-oriented new built-up land extraction method using multi-temporal Sentinel data and ensemble learning algorithm is proposed,and the distribution characteristics of new built-up land in Nanjing from 2016 to 2019 are analyzed.The main research contents and conclusions of this paper are as follows:(1)A hybrid feature set of multi-temporal Sentinel-1A and Sentinel-2A for new built-up land extraction is constructed.Based on multi-temporal Sentinel-1A and Sentinel-2A data,21-D features,including spectral features(blu-ray reflectance,green light reflectance,red light reflectance,near-infrared reflectance,NDVI,EVI,NDWI and soil brightness,green veg and yellow stuff from KT transformations),texture features(the first principal component of Principal component Analysis and its' 8 texture operators by calculating Gray Level Co-occurrence Matrix,including mean,variance,homogeneity,contrast,dissimilarity,entropy,second moment and correlation)and backscatter features(VH and VV)are extracted,and the spectral and spatial characters of various land cover types including built-up are analyzed.Then,object-oriented segmentation,global mean filtering and normalization are carried out,and the dataset used for new built-up land extraction is constructed by combining two phase's feature set.(2)A new built-up land extraction method based on ensemble learning is proposed.Based on multiple classifiers,including random forest,rotation forest,support vector machine and extreme learning machine,ensemble learning algorithm is constructed by precision weighted voting method,and used to extract new built-up land of dataset by binary classification.Comparing the extraction accuracy of base classifiers and ensemble learning,and comparing the extraction effect of proposed direct change extraction strategy and post-classification comparison strategy,it is found that the extraction effect of direct change extraction strategy based on ensemble learning is optimal,better than single classifier or post-classification comparison method,with it's overall accuracy up to 0.95 and Kappa coefficient up to 0.89.The pepper and salt phenomenon of extraction result are weaker,the shapes are more complete and the contour is clearer.(3)This paper analyzes the quantity,spatial layout and changes in the ecological Red Line area of the new built-up land in Nanjing from 2016 to 2019.Based on Nanjing new built-up land extraction results,we census new built-up land's area of the city and the administrative districts from 2016 to 2019,new built-up land's area of 2016-2017 is 66.3191 square kilometer,of 2017-2018 is 54.7966,of 2018-2019 is 38.6442,while the total of 2016-2019 is 153.5371 square kilometer,accounting for 2.3184%of the city's size.The geometric center of gravity of the new built-up land for each administrative district in each year are generated,the center of is distributed in the upper part of the Jiangning District,and the center of the new built-up land moves northward first and southward afterwards,but it is always distributed near the center of the city.Combining with the scope of ecological Red Line area,we find that the area of new built-up land found inside ecological red Line area of 2016-2017 is 4.8481 square kilometer,of 2017-2018 is 2.7464,of 2018-2019 is 2.8398,while the total of 2016-2019 is 10.0893 square kilometer,accounting for 6.5712%of the total area of new built-up land and 0.1523%of Nanjing.Total area inside the first level control area is 0.668 square kilometer,most are distributed in the secondary control area.