Identification Of Grassland Degradation Indicator Species Based On UAV Hyperspectral Remote Sensing

Grassland,as the supply bank of material needed by human beings,has been degraded in different degrees in recent years with the rapid development of social economy and population increase.The identification and quantitative statistics of grassland degradation indicator species are the basis of grassland monitoring and grassland management policies implemented by local governments.UAV Hyperspectrum has been applied to grassland monitoring due to its characteristics of integrated spectrum,strong mobility,great flexibility,and high spatial resolution and hyperspectral resolution of images obtained.Based on typical Qing Hai-Tibet plateau peat bog Zoige plateau distribution area as the research area,in Unmanned Aerial Vehicle(UAV)Hyperspectral data and feature spectrometer based on the measured data,combined with field investigation,through different hyperspectral indicator species identification method completes the indicator species of grassland degradation of regional identification,and through comparison,verify the identification precision of the two methods,The differences of indicator species in different degradation gradients were analyzed to provide scientific basis for remote sensing monitoring of grassland degradation in this area.The main research results of this paper are as follows:(1)Construction of standard ground object spectral database.In this paper,through the field measurement of handheld spectrometer,View Spec Pro software was used to complete the calculation of its average spectrum and a series of subsequent processing,and finally completed the establishment of standard spectral data of 12 sample sites.The results showed that the reflectance of spectral curves of the same species in different degraded plots was different,so different grassland species could be classified based on the characteristics of spectral curves.(2)End element extraction of hyperspectral image.In this paper,three methods of Vertex Component Analysis(VCA),Orthogonality Subspace Projection(OSP)and Continuous Maximum Conical Cone(SMACC)were used to extract end-member spectra from hyperspectral images of 12 plots.The results showed that Vertex Component Analysis(VCA)had a strong ability to extract end-member spectra of vegetation in swamp soil,peat soil and meadow soil samples when the number of species was large.When the number of species was small,the extraction capacity of the three methods was basically the same.However,Orthonormal Subspace Projection(OSP)has a strong ability to extract endmembers from hyperspectral images in severely degraded desertification plots.(3)Three end member spectral matching techniques.In this paper,Spectral Angle Matching(SAM),Spectral Feature Fitting(SFF)and Binary Encoding(BE)are used to compare and analyze the image spectrum with the spectrum in the spectral library.Finally,the category of each image end member spectrum is determined by weighted assignment method.The results showed that the three spectral matching methods were more suitable for the dominant species of grassland in the 12 sample plots,and the order was Binary Encoding(BE)> Spectral Angle Matching(SAM)>Spectral Feature Fitting(SFF)from high to low.(4)Two classification methods and accuracy.Spectral Angle Mapping(SAM)and Spectral Information Divergence(SID)were used to complete the spatial distribution map of grassland species in this region,and the results were verified.The results showed that: from the classification accuracy of dominant species in grassland in this region,the classification accuracy based on complete waveform was the highest,with the classification accuracy ranging from 61% to 100%,the overall classification accuracy ranging from 47.73% to 87.5%,and the Kappa coefficient ranging from 26.7% to 81.4%.The classification method of Spectral Angle Mapping(SAM)is more suitable for grassland species identification in this area.