Research On Forest Information Extraction In Northeast China Based On Multi-source Remote Sensing Data

Forests are the largest ecosystems on land,regulating the global carbon and water cycles.Forest resources are important natural resources,and their horizontal distribution,vertical structure,and storage volume are the basic research objects of statistical forest resources.Remote sensing technology has the advantages of high efficiency,low cost and large-area simultaneous observation,and has gradually developed into a main method of forest information resource investigation.With the continuous enrichment of satellite data,the use of synthetic aperture radar(SAR)data and optical remote sensing images can achieve more accurate extraction of forest structure information from different levels.On the basis of summarizing the research on the extraction of forest parameters at home and abroad,this paper selects Jingyuetan National Forest Park in Changchun City,Jilin Province as the experimental area,and comprehensively uses SAR data and multispectral remote sensing data to carry out forest information extraction research.The main contents are as follows:(1)Research on Forest Type Recognition Method Based on Multi-Dimensional PolSAR.Taking GF-3 and ALOS-1 PALSAR remote sensing images of different phases as data sources,combining the advantages of different sensors and different types of SAR data,three polarization characteristics of entropy,scattering angle and anisotropy are obtained by decomposition based on scattering mechanism.Then based on the gray level co-occurrence matrix,three texture features of mean,variance and heterogeneity are extracted.The above feature parameters are introduced into the support vector machine(SVM)classifier,and multi-dimensional SAR information such as multi-band,multi-polarization,multi-temporal and texture features is comprehensively used for forest type identification.The overall classification accuracy of the experimental results is 89.47%,and the Kappa coefficient is 0.85.(2)Research on forest type recognition method based on SAR and optical data.Using Sentinel-1 SAR image and Sentinel-2 multispectral image as data sources,the backscattering coefficient map is obtained by polarizing the SAR data,and two texture features of mean and entropy are extracted.Combined with the red edge band and the spectral vegetation index characteristics of the optical data,the random forest classification method is used to realize the identification of the forest type in the experimental area.The overall classification accuracy reaches 92.54%,which is higher than the classification accuracy using only SAR data(63.12%)or the classification accuracy of optical data(84.07%),indicating that the fusion of SAR and optical features can effectively improve the recognition accuracy of forest types.(3)Research on Forest Height Inversion Method Based on Multi-source Remote Sensing Data.Based on Sentinel-1 SAR image,using Polarization and Interference(InSAR)technology,the VV/VH polarization backscatter coefficient,angle of incidence,phase,coherence,and digital elevation model(DEM)of the experimental area are obtained,and slope information is extracted from the DEM.Then calculate the spectral vegetation index and biophysical variables based on the Sentinel-2multispectral image,and use the XGBoost algorithm to connect the ground measured tree height data with the SAR polarization information,interference information,spectral vegetation index,biophysical variables and terrain factor,so that multi-source remote sensing data can be used to retrieve the forest height.The experimental results show that the VH cross-polarization characteristic parameters obtained based on SAR data are more conducive to tree height inversion than VV co-polarization characteristic parameters,indicating that cross-polarization can better detect forest structure information.The inversion model that combines optical characteristics and VH polarization characteristics of SAR data has the best estimation accuracy,and the determination coefficient(~2R)and root mean square error(RMSE)are 0.757 and 1.648respectively.This method solves the problems of inaccurate ground phase estimation,high algorithm complexity and high cost of lidar data in the existing forest height estimation algorithm.Combined with the respective advantages of SAR and optical data,this paper uses multi-source remote sensing data to extract feature variables related to forest structure information,and uses machine learning classification and regression algorithms to achieve forest type identification and forest height inversion methods in Northeast China.