Extraction Of Impervious Surface Information In Nanchang City Based On Multi-source Remote Sensing Images

One of the important indicators for measuring the degree of urbanization and evaluating the quality of urban ecological environment is impervious surface.As one of the typical representative regional cities of Poyang Lake Ecological Economic Circle,accurate and efficient acquisition of impervious surface information of Nanchang is of great significance to complete the Poyang Lake Ecological City Cluster Plan(2015-2030).Most of the existing impervious surface studies focus on extracting impervious surfaces from a single optical remote sensing image,but due to the influence of image data sources,the classification results often suffer from spectral confusion and "pretzel" phenomenon,which limits the further improvement of impervious surface classification accuracy.In order to improve this situation,this paper takes Nanchang City as the study area,and explores the applicability and advantages and disadvantages of extracting impervious surface information based on multi-source remote sensing images by considering the characteristic advantages of multi-source remote sensing data.The main findings of the experiment are as follows.(1)The study compared the impervious surface classification accuracy of four machine learning algorithms using the Lantsat-8 multispectral band.The experimental results show that random forest has higher classification accuracy.(2)The study extracted the impervious surface information of Nanchang City using the feature information of dual-polarized SAR data and evaluated the accuracy.The experimental results showed that the overall classification accuracy,Kappa coefficient and impervious surface extraction accuracy were improved by 5.86%,0.082 and 14.55%,respectively,comparing with the classification results of Lantsat-8 multispectral band,and the classification effect was better.(3)The experiment fused Landsat-8 data with Sentinel-1A data at the feature level and extracted impervious surface information from the Nanchang City area.The experimental results show that the overall classification accuracy,Kappa coefficient and impervious surface extraction accuracy of fused multi-source images are improved by 16.02%,0.206 and 24.73%,respectively,compared with the classification results of Lantsat-8 multispectral bands.Compared with the classification results of a single image,the combination of feature information from multiple sources can effectively solve the feature confusion problem and the "salt and pepper" phenomenon,and the classification results have significantly improved image quality.(4)The corresponding classification models were constructed according to different feature combinations,and the comparison of the impervious surface extraction results of different classification models showed that the texture feature information of Sentinel-1A images significantly improved the impervious surface classification accuracy,followed by the backward scattering feature information,while the texture features of Landsat-8 images were poorly improved.(5)The study obtained time-series impervious surface information of Nanchang City by fusing multi-source remote sensing images,and analyzed its spatial and temporal dynamic characteristics and driving factors.The experimental results show that from 2014 to 2019,the impervious surface of Nanchang City maintains a rapid growth in general,and is concentrated on both sides of Ganjiang River in a "faceted" manner,with the main axis of impervious surface expansion to the west.The results of the driving force analysis combined with the yearbook data and information inquiries show that economic development,population size and policy regulation are the main drivers of impervious surface,where the correlation coefficients of economy,population and impervious surface area are 0.957 and0.977,respectively.