Campus Land Use Classification By Integrating Remote Sensing And Social Media Data

Land use classification data is the basic data for geographies and related subjects.The acquisition of high-precision land use classification data is a current research hotspot Through the satellite remote sensing imagery with the high spatial and temporal resolution,rapid classification of land cover has been solved.However,the remote sensing imagery can only reflect the "surface" features,and it is difficult to characterize the inherent human-land relationship,so obtaining high-precision land use classification still faces challenges.At present,several studies to carry out high-resolution classification of urban land use combined with social big data,and interesting results have been achieved.As a core area of talents in the city,the precise classification of land use on university campuses based on social big data has few reports.Based on this,the study collected multi-source data from the campus area,including GF-2 campus imagery,Open Street Map[1],and Wi-Fi login data.First,the GF-2 multi-spectral image and panchromatic image are used to generate a campus multi-spectral image with high spatial resolution.The object-oriented image classification method is used to generate campus land cover classification map.Combined with the characteristics of campus students and staff mobility,the rules extracting from campus Wi-Fi login data were formulated.Based on this data,machine learning methods such as C4.5 algorithm,random forest,multi-layer perceptron,K-means algorithm,and ISODATA clustering are used to identify campus building functions.And then defines the cluster meaning with POI data that download form OSM.By comparing the accuracy of campus building recognition results,the classification algorithm with the highest accuracy—random forest algorithm is selected to generate a campus building function classification map.Finally,the ArcGIS overlay analysis method is used to generate a high-resolution campus land use classification map.In the process of campus building function identification,it is found that the traditional distance-based K-means algorithm which classification accuracy is 46.5%and ISODATA clustering which classification accuracy is 64.4%have low classification accuracy.The supervised algorithms such as decision tree C4.5 algorithm which classification accuracy is 77.5%,random forest which classification accuracy is 85.0%,and multi-layer perceptron which classification is 80.0%have better results.Experiments show that Wi-Fi login data is a reliable data source that can effectively identify campus building functions.All in all,this research not only enriches the theory and methods of Wi-Fi login data in the campus land use classification but also provides data support for campus land management and monitoring.