Remote sensing recognition of residential areas based on GF-4 satellite image

Abstract

Residential area is an important place for human habitation and life. Using remote sensing technology to identify residential areas is of great value for land resources planning and utilization, disaster prevention and relief and other fields. As the world's first high resolution optical remote sensing satellite for geosynchronous orbit, GF-4 satellite has high time resolution, medium spatial resolution and multispectral land detection capability, which provides a new data resource for residential monitoring. Closely combining with the detection characteristics of the GF-4 satellite, this paper proposes a remote sensing identification method based on GF-4 satellite, and the recognition ability of the GF-4 satellite to the residential area is analyzed. The remote sensing recognition of residential areas is mainly divided into four steps. First, Super-resolution image enhancement technology is used to improve the spatial resolution of GF-4 satellite PMS image. Then, the resolution enhanced image is processed by geometric correction, radiometric calibration and atmospheric correction. Third, the existing land use and land cover data are selected as prior knowledge to select typical sample areas. Based on the spectral characteristics and spectral relationship of different objects in GF-4 satellite image, decision tree classification method is used to eliminate the obvious non-residential areas such as cloud, vegetation, water and shadow, so as to reduce the subsequent data processing and reduce the false recognition rate in residential area. Finally, SVM classifier is selected for the classification of residential areas. Taking GF-4 satellite data in Jiashan county as experiment, the result shows that the user accuracy of resident recognition by this method is 89.96%, which is significantly higher than that without resolution enhancement in the same method. Besides, the spatial scope of the county and township residents can be effectively identified in GF-4 enhanced image.