Study On Urban Land Use Classification Combining Remote Sensing And Spatial Big Data

Urban land use is an important premise and basis for implementing urban land use policy and controlling urban land development.It is of great significance for mastering urban structure,solving urban problems and carrying out urban planning scientifically and reasonably.Remote sensing technology has been widely applied in land use classification and dynamic monitoring of surface change due to its rapid development.However,the types of land use within cities show extremely high similarities in remote sensing images,and they often have certain functional and semantic attributes.It is a challenge to achieve land use classification relying on remote sensing images in urban areas.Powered by advanced communication technology,the incessant emerging of spatial big data with temporal and spatial semantic information provides new data sources for urban land use classification.Also,the update of data mining technology provides technical method.Nevertheless,the existed studies have shortcomings in terms of process and recognition scale.The classification process in most studies is cumbersome.And only the dominant functions of different urban land use at the block scale were identified,which is difficult to meet the needs of fine urban management.Remote sensing images could reflect the physical attributes of different land use types,while spatial big data could represent the semantic functional information.The combination of multi-source data provides a new perspective to achieve accurate urban land use classification.Based on multi-source remote sensing data(Sentinel-2A,Sentinel-1A and Luojia-1 nightlight images)and spatial big data(points of interest,population fever data),this paper extracted and analyzed the features of multi-source data.Combining with the random forest model,the accurate classification of land use in the main area of Lanzhou has been achieved.Meanwhile,a series of feature combination classification experiment was designed to evaluate the contribution and influence of different features on the classification results.Furthermore,the distribution characteristics of urban land use in Lanzhou was analyzed.The conclusions of this article are as follows:(1)Both the Place2Vec model and the Word2Vec model could extract the semantic features of POI data.Compared with the Word2Vec model,the Place2Vec model could fully mine the potential semantic and spatial information of POI,and it has a better classification effect when applied to urban land use classification.(3)The synthetic application of multi-source features is conducive to the improvement of urban land use classification.The urban land use classification model that integrates remote sensing and spatial big data has achieved good classification results with overall accuracy and the Kappa coefficient of 83.00% and 0.75 for level?.Base on spectral and texture features,the addition of backscatter and night light information could improve the classification accuracy of some land categories,while POI and population thermal features could improve accuracy of all classes significantly.(3)The importance of different types of features was analyzed.The results showed that population thermal information at a single moment contributes little to the classification.In contrast,the population thermal at multiple moments has the largest contributions of all features.The contributions of spectral and POI features are small relatively,and night light and backscattering features have the weakest combinations.(4)Analyzing the spatial distribution characteristics of urban land use in Lanzhou,it is found that the spatial distribution of different types of urban land has obvious differences.Industrial and residential lands are the major land types as a whole.Specifically,the industrial lands cover the largest area,showing the characteristics of large area concentrated and continuous distribution,mostly in Xigu District and the periphery of Chengguan District.The residential lands cover the second largest area after the industrial land,mainly distributed in the central area of Chengguan District and the eastern area of Qilihe District.The commercial lands cover the smallest area,which locate more in the east and south areas,less in the west and north areas.Except to undeveloped areas,the public lands are generally more fragmented,concentrated in Chengguan District and Anning District.