| With the development of human activities and the world,the forest resources are greatly reduced,and the forest cover changes rapidly.People need to obtain the information of forest cover changes quickly and accurately.Sentinel remote sensing satellite contains radar data and optical remote sensing data.Because of its short revisit period and rich band information,it has high research value in forest land cover change monitoring.As the "Pearl of Dongting",Yuanjiang City is very important to the ecological environment of Dongting Lake.However,due to the influence of natural environment,human interference and policy factors in recent years,the natural features of Yuanjiang City have changed greatly,so it is of great significance to explore the changes of the natural features and forest land cover in Yuanjiang City.In this paper,Yuanjiang City is taken as the research area.Because Sentinel-la satellite revisit cycle is 12 days,Sentinel-2a revisit cycle is 10 days,the timeliness is strong,the data can be obtained free of charge,and has high use value,sentinel data is selected as the research data source.Using the spectral index,texture characteristics,backscatter coefficient of Sentinel remote sensing data,combined with the field survey data and historical data,to explore the accurate and efficient classification method of Yuanjiang City.Select the appropriate land feature classification method to classify the Sentinel image data of Yuanjiang City in July 2016,May 2017 and July 2018,and analyze the change and trend of forest land cover in Yuanjiang City.The main research results and conclusions are as follows:(1)Combined with the field survey data,using Sentinel-2a image data,select the training samples of Yuanjiang City’s surface features,analyze the spectral and texture characteristics of each surface feature,fully mine information,explore the rules of distinguishing between different objects,build decision tree,and accurately classify the types of objects in the study area.After comparing the accuracy of classification results based on decision tree with that of supervised classification based on maximum likelihood method,it can be found that the overall classification accuracy of Yuanjiang City Based on knowledge decision tree is 11.27%higher than that of supervised classification based on maximum likelihood method,and the overall kappa coefficient is 0.133 higher.Using Sentinel-2a image data,combined with spectral index and texture features,the knowledge decision tree classification method is more accurate and effective than the maximum likelihood method.(2)The fusion of Sentinel-1a data and Sentinel-2a data can improve the classification accuracy.After resampling the data in VV/VH polarization mode of Sentinel-la data and fusing it with Sentinel-2a data,the backscattering coefficient of ground objects is introduced,and the classification results are analyzed together with the spectral index and texture features.The comparison between the classification results of Sentinel-2a data and that of Sentinel-2a data alone shows that after fusing the data,the ground objects are classified,which is more based on the decision tree of multi features than Sentinel-2a data alone The overall classification accuracy of the class increased by 6.13%,and the overall kappa coefficient increased by 0.0504.(3)From July 2016 to May 2017,the total area of forest land in Yuanjiang City decreased by 849 hectares.In 2002,the forest land became other features,and 1153 hectares of other features became forest land.Between May 2017 and July 2018,the total area of forest land decreased by 304 hectares,875 hectares of forest land became other features,and 571 hectares of other features became forest land.Forest land is mainly converted into cultivated land and construction land,so on the premise of ensuring the production and living of residents and urban construction,it is also necessary to strengthen the protection of forest land resources.The spectral index and texture features selected in this paper are limited.In the future research,we can further explore whether other spectral index and texture features can improve the classification accuracy.In the analysis of forest land cover changes,only the mutual changes between forest land and other features are explored,and factors such as topography and tree species can be added to analyze. |
