Land Cover Classification Method Based On Multi-source Remote Sensing Data

Land cover classification is an important research direction in the field of remote sensing.It plays an important role in the basic research of environmental science,sustainable management of land resources,and the relationship between people and nature.Land cover information is closely related to human production and life,and is a basic element that affects and reflects the relationship between human beings and the natural environment.The change of land cover reflects the process of nature,ecological environment and human society,and affects climate and biological diversity on a global scale.The current activities of human society are changing the natural environment at an unprecedented rate.Changes in land cover caused by human social activities are important factors in the current global land cover change.Accurate land cover classification information is very important.Rapid access to a wide range of high-quality land cover classification maps remains a challenge.In order to improve the quality of land cover classification and mapping,this study makes full use of multi-source remote sensing data to construct a comprehensive feature space,including spatial features(such as texture,shape,area,etc.),temporal features(time series features)and spectral features.A total of 63 features were constructed in time,space and spectral dimensions,and the advantages of multi-source remote sensing data were considered more comprehensively.A wealth of feature information has been established.Before the feature extraction,the multi-spectral data was subjected to atmospheric processing,resampling,cloud removal,cropping and splicing.The data processing of radiation correction,terrain correction,speckle filtering,cropping and splicing is performed on the synthetic aperture radar data.To reduce the weather,atmospheric particle scattering and sensor sensitivity interference of remote sensing data received.An object-oriented land cover classification method based on hierarchical structure is proposed.Refine the classification into a multi-level sub-category problem.One class is distinguished at each level.In each layer,appropriate segmentation parameters and feature subsets are dynamically selected for each type of feature to obtain high integrity and high precision classification results.Finally,the prototype system of this method was implemented and verified by Beijing area as the research area.Based on the three classifiers,the single-scale classification method and the method proposed in this thesis are used to produce land cover classification maps for Beijing.The method proposed in this thesis achieves a total accuracy of 91.6%,and the overall accuracy of the single-scale classification method is improved by 2.43%,0.55% and 1.58%,respectively.