Forest Land Remote Sensing Classification And Change Monitoring ——A Case Study Of Genhe City

In recent years,China’s remote sensing technology has achieved rapid development,providing strong support for the development of China’s environmental protection and forestry informatization.Thereinto,the use of multi-temporal medium and high spatial resolution remote sensing data to carry out large-scale classification of forest types and to monitor the dynamic changes of forest resources has become one of the research hotspots.In this paper,for the purpose of forest resources monitoring and investigation and forest environment protection,based on the latest WFV（Wide Field of View）data from the domestic satellite Gaofen-6,an evaluation study on geometric accuracy and classification accuracy is carried out,and the application potential of the newly added four bands in forest land type classification research is tested.Based on the data series of domestic Gaofen Satellite（hereinafter referred to as GF）and Landsat series data,the research on forest land information extraction and change monitoring is conducted,and the change causes are analyzed,which provides reference for the spatial allocation of forest resources,environmental protection and the application of domestic latest GF data.The main contents and conclusions of this paper are as follows:（1）The geometric accuracy and classification accuracy of GF WFV 1A data were tested.The geometric accuracy of images was tested by rational polynomial coefficient model.The classification accuracy was tested by random forest classification method using 6 different band combinations.The results show that the geometric distortion of GF-6 WFV data is slight,the median error of image registration after uncontrolled RPC correction is within 10 meters,the newly added red edge 1,red edge 2 and yellow edge wave segments have significant influence on forest type classification,the overall classification accuracy is increased from 74% to more than 80%,and the purple edge band also contributes to forest type classification.（2）Taking Genhe City,Hulun Buir in Inner Mongolia as the research area,with the support of multi-source data,three machine learning algorithms（maximum likelihood method,support vector machine and random forest）are adopted.The spectral features,texture features and terrain features of the images are comprehensively utilized,and the forest resource planning and design survey results are combined with classified samples and verification sample data obtained from GF-1PMS and GF-2 high-resolution images.According to the accuracy verification results,an effective method for extracting forest type information in the research area is selected.The results show that among the three machine learning algorithms of maximum likelihood,support vector machine and random forest,random forest algorithm has the highest classification accuracy for this region.（3）The 4 phases of images were classified into 6 categories by post-classification comparison method.They are coniferous forest,broad-leaved forest,shrub forest,burned area,water body and others（mainly urban land and cultivated land）.A total of4 phases of land use classification maps were obtained.The areas of coniferous forests,broad-leaved forests,shrubbery and burned areas are counted,and the change trend and reasons are analyzed.The results show that coniferous forest area accounts for the highest proportion in Genhe City,and the area of forested land（coniferous forest,broad-leaved forest and shrub forest）shows a slight downward trend from 2009 to 2018.The area of coniferous forest and shrub forest shows a downward trend and the area of broad-leaved forest shows an upward trend.The analysis of the change factors mainly includes the fire factor and meteorological factors,which are the results of the combined action of the two.