Extraction Of Urban Impervious Surface And Its Spatio-temporal Evolution Based On Landsat Imagery

Urbanization refers to the process of the land transformation from rural to urban with the adjustment of regional industrial structure,and population growth and the transformation of land use types.Impervious surface change can be used as an indicator to measure the level of urbanization.With the development of remote sensing technology,a large number of multi-temporal detection datasets are aquired for the study of the temporal and spatial evolution of urban impervious surface.Landsat images were taken as the data source,and the main urban area of Chengdu was selected as the study area.The research work on impervious surface carried out on Landsat images in 2001,2005,2009,2013 and 2017 includes the following three points:(1)Based on Normalized Difference Vegetation Index(NDVI),Modified Normalized Difference Water Index(MNDWI)and Biophysical Composition Index(BCI),the interpretation of vegetation,impervious surface,bare land and water body in remote sensing images were constructed as sample data sets.(2)Based on the samples from May 1,2017,construct a BP neural network model based on one-dimensional data and a convolutional neural network model based on two-dimensional data,and explore the best impervious surface extraction algorithm.The experimental results showed that the BP mistakenly classified the impervious surface as bare land,with an overall accuracy of 83.31%,Kappa coefficient of 0.78,and Macro F1 value of 81.27%;the convolutional neural network did not mistake the impervious surface as bare land.The overall accuracy is 98.32%,the Kappa coefficient is 0.98,and the Macro F1 value is 98.28%.Compared with the BP,the convolutional neural network has improved 15.01%,0.20,and 17.01% in Macro F1,overall accuracy,and Kappa coefficient,respectively.(3)With remote sensing data of October 28,2001,April 14,2005,and March 24,2009,April 12,2013,and May 1,2017,the most effective CNN structure was used to extract the impervious surface information on the images to explore the spatial and temporal evolution characteristics of impervious surface in the past 16 years.The experimental results showed that:(1)The impervious surface is the largest land use/cover type in the urban districts of Chengdu.The increase area of impervious surface from 2001 to 2009 mostly comes from the vegetation area.From 2009 to 2017,the main changes of impervious surface Occurred in the process of urban demolition and construction;(2)From 2001 to 2009,the urban areas of Chengdu mainly changed the previous vegetation area into impervious surface.After 2009,the impervious surface area of Qingyang District and Jinniu District was almost unchanged,there are urban demolition and reconstruction processes in Wuhou District,Chenghua District and Jinjiang District.(3)In 2001,2005,2009,2013 and 2017,the central point of the impervious surface in urban District of Chengdu remained almost unchanged,and all were distributed in the Qingyang District.(4)In 2001,2005,2009,2013 and 2017,the impervious surface of urban District of Chengdu has no obvious distribution direction.