Research On Urban Land Extraction Algorithm Based On SVM

China has entered the stage of rapid urbanization.Dynamic monitoring and accurate predicting of urban expansion has become the focus of academic research and government decision-making priorities.The global nighttime light data obtained from the Operational Linescan System(OLS)sensor carried by the US Military Meteorological Satellite(DMSP)is an effective data source for large-scale urbanization studies.The extraction of urban land use information by DMSP / OLS nighttime light data can provide scientific basis for the study of spatial pattern distribution of large-scale urban land use.And the effective image recognition classification algorithm is the key to the extraction of urban land.Combining with the improved support vector machine classification algorithm,the urban land in nighttime light data is extracted,and the spatial and temporal pattern of urban land use is analyzed and the future expansion scale is simulated.To provide theoretical basis and reference for the development of macroeconomic policies.This paper used the invariant target region method to correct the long time sequence DMSP / OLS nighttime light data.We proposed an improved SVM classification algorithm to extract the boundaries of representative towns in Jiangsu Province.Then,we analysis the spatial expansion pattern,overall development trend and center shift.Finally,GM(1,1)gray model was used to predict the future development of Nanjing.The main conclusions of this paper are as follows:(1)We used the invariant target area method to correct the nighttime light data's saturation effect,which can effectively alleviate the saturation effect of light data;Through continuity correction and pixel anomaly fluctuation correction,reducing the difference in sensor image for the same year.Through reducing the fluctuation of data in adjacent years,we can improve the continuity and comparability between data in different years.(2)We proposed an improved SVM classification algorithm to extract urban land,combined with Landsat 8 image by artificial extraction of urban land for precision evaluation.The results of the comparison showed that the improved SVM classification algorithm proposed in this paper has the advantages of overall accuracy,Kappa coefficient and user accuracy.(3)Based on the fan-shaped analysis,concentric circle analysis and average transfer of urban land center,we found that Nanjing has experienced the slow and rapid expansion process from the long-term sequence.Nanjing spread to the surrounding area on the basis of the main city in 1992,along the Yangtze River and north-south traffic corridor development.Urban land density continues to increase from 1992 to 2013.(4)Finally,we used GM(1,1)grey model to forecast the future development of Nanjing.The accuracy of the GM(1,1)model was calculated: the mean square error index C = 0.1745,the small error probability P = 1.0000.The mean square error index C <0.35,the small probability error P> 0.95,indicating that the model accuracy is very feasible.It is found that the forecasting scale of urban land use in Nanjing in 2018 and 2020 is 2030.88km~2 and 2564.26 km~2,respectively.