Vegetation Classification In The Typical Region Of The Yellow River Source Based On HSI Hyperspectral Remote Sensing Data

Vegetation plays an important role in reflecting changes in the ecological environment.In recent years,it has gradually become one of the hotspots in the field of ecosystem environment research.Among them,the vegetation classification has extremely important significance for the study of vegetation dynamics and vegetation coverage.The traditional vegetation classification needs to go to the field for field investigation,which is not only time-consuming and labor-intensive but also not suitable for large-scale operations.The rise of remote sensing technology provides an efficient way to obtain vegetation information,identify and classify vegetation types.However,because of its discontinuous wave band,multi-spectral remote sensing can only distinguish the types of vegetation with large spectral differences,but it is not obvious for the classification of vegetation with small spectral differences.Compared with multi-spectral remote sensing,hyper-spectral remote sensing provides a brand-new technical means for the field of vegetation classification due to its obvious characteristics of high spectral resolution,large number of bands,narrow band width and continuous.This paper selects the western and northern Eling Lake as the main research area in the core area of the Yellow River source area of the Three Rivers Source National Park in the Qinghai-Tibet Plateau.First,based on the Indian Pines hyperspectral data set,the spectral angle is used to replace the Euclidean distance modified ISOMAP and LLE fusion algorithm(SA_LLEISOMAP),Feature extraction of pre-processed hyperspectral images,and five other nonlinear dimensionality reduction algorithms: local linear embedding LLE,isometric mapping ISOMAP,fusion algorithm LLEISOMAP,spectral angle isometric mapping SA_ISOMAP and nuclear principal component analysis KPCA)By comparison,the superiority of the SA_LLEISOMAP dimensionality reduction algorithm is verified by comparing the average gradient,edge strength,information entropy,and regularized feature map.Afterwards,three commonly used supervised classification algorithms(support vector machine,maximum likelihood,random forest)were used to classify the vegetation of HJ-1A HSI hyperspectral image after the dimension reduction in western Eling Lake,and the classification results were compared and analyzed.Completed the production of vegetation classification map in the west of Eling Lake.Finally,based on the same classification algorithm,the vegetation classification map of northern Eling Lake was completed.The vegetation classification maps in the west and north of Eling Lake obtained in this paper have a certain reference value for vegetation monitoring and protection in the core area of the Yellow River Park in the Sanjiangyuan National Park.