A Study On Classification Of Typical Vegetation Communities In Xianghai Wetland Using HSI Hyperspectral Remote Sensing

Feature extraction and recognition precision based on hyperspectral remote sensing data is the research hotspot issues in the field of remote sensing currently;in this paper,using RS(remote sensing,RS)and GIS(Geographic Information the System,GIS)as technical support,selecting Xianghai Nature National Reserve as the study area,select the Phragmites australis,Typha orientalis,Suaeda glauca and Leymus chinensis as the research object,application of 2015 years HJ-1A/HSI hyperspectral remote sensing,combined with the vegetation hyperspectral reflectance data of the measured ground,using spectral angle mapping and spectral information divergence classification method classify the typical vegetation communities in the study area.Firstly,smoothing the four typical vegetation community reflectance data,used to establish vegetation endmember spectra library;secondly,on the basis of pre-proceeding image,compared the principal component analysis method and the minimum noise separation method and dimensionality reduction and feature band selection for hyperspectral image denoising;finally,with the processed data,use First derivative analysis and continuum removal analysis,highlight the spectral reflectance and absorption characteristics of the typical vegetation,highlight the different spectral differences between different types of vegetation,to guide the SAM and SID methods to identify the typical vegetation community classification.The main research results are as follows:(1)HSI hyperspectral data there are obvious oblique stripes,the first 20 bands is particularly evident,but the overall number less.The image data after the banding can truly reflect the spectral characteristics of the surface vegetation and can be used to identify the vegetation community.(2)The reflectivity data of the typical vegetation community measured by the ground has the spectral characteristics of the "two peaks and three grains" of typical healthy green plants.Using spectral differentiation method,it is possible to highlight the difference of absorption characteristics between typical vegetation community types,which is helpful for high precision recognition.At 458.4nm,Phragmites australis and Leymus chinensis easier to distinguish but Typha orientalis and Suaeda glauca is not easy to distinguish;at 546.2nm,Typha orientalis and Suaeda glauca is easy to distinguish;at 609.5nm,Typha orientalis and Suaeda glauca is easy to distinguish but Typha orientalis and Leymus chinensis is not easy distinguished;at 820.5nm,Phragmites australis,Suaeda glauca easily distinguish with Leymus chinensis;at 904 nm,the Typha orientalis,Suaeda glauca is easy to distinguish with Leymus chinensis;continuum removed after analysis,at 576 nm,647 nm,898 nm,920nm,can accurate identification the typical vegetation communities.(3)Classification results showed that: after MNF converting denoised characteristic image,using SAM method classifies overall accuracy slightly SID classification result of the process,the overall classification accuracy is 89.0854% and 85.4901%,the Kappa coefficient was 0.8058 and 0.7983.In summary,through transformation and analysis of hyperspectral data of actual measured typical vegetation communities on the ground,with the HJ-1A/HSI hyperspectral data,analyed of the differences between the spectra characteristics of typical vegetation,the image preprocessing could reflect the spectral characteristics of typical vegetation communities in Xianghai wetland,and the SAM classification method could obtain high classification accuracy.The This paper provides a reference for the study of hyperspectral identification of vegetation communities in arid and semiarid regions,so as to improve the algorithm of wetland vegetation information extraction.It is very important to expand the domestic hyperspectral remote sensing image in the application of fine identification of vegetation.