Vegetation Classification Of Alpine Grassland In Qinghai Lake Basin Based On HJ-1A Hyperspectral Data

The Qinghai Lake Basin is located in the northeastern part of the Qinghai-Tibet Plateau.It is a typical arid area with extremely fragile ecosystems.The type and distribution of vegetation in the territory has always been a problem for research.However,the altitude of the area is generally higher(above 3200m),the area is larger(2.96×104km2)and the vegetation types are diverse.Compared with the traditional field sample survey and multi-spectral visual interpretation,the HJ-1A HSI hyperspectral remote sensing data because its wide coverage,many bands and rich spectral information have significant advantages in the fine classification of vegetation in this area.This paper takes into account the research and practical application requirements.Firstly,the pre-processed single-view HSI hyperspectral remote sensing data of Qinghai Lake Basin is used to test,and the results of various dimensionality reduction and classification algorithms are compared to obtain the suitable processing method and process for HSI hyperspectral remote sensing data.And then applied it to the 31-view HSI data covered entire basin.Finally,the production of a 1:500,000 vegetation type map was completed.The test of single-view HSI data includes: Firstly,three kinds of dimension reduction are MNF,KPCA and ISOMAP are used to reduce the dimension of the image.By analyzing the regularized feature map of the reduced-dimensional image,it is concluded that ISOMAP has obvious advantages in the information concentration when processing HSI data.After comparing the advantages and disadvantages of ISOMAP based on traditional Euclidean distance and IMED-ISOMAP based on image Euclidean distance,it is considered that there is little difference between information entropy and average gradient.However,the ISOMAP algorithm based on the traditional Euclidean distance is more efficient and is suitable for the dimension reduction application of HSI images in Qinghai Lake Basin.The reduced-dimensional image is classified by three classification algorithms: maximum likelihood,artificial neural network and support vector machine,and then its accuracy is compared.It is concluded that SVM is superior to the other two in hyperspectral data classification.The results of single-view HSI hyperspectral remote sensing data show that ISOMAP+SVM is more suitable for the classification of grassland vegetation in Qinghai Lake Basin.Based on the above conclusions from the test on single-view image,the ISOMAP dimension reduction and SVM classification methods are combined to process the HSI data of the 31-view images in the Qinghai Lake Basin,supplemented by field surveys and high spatial resolution images.The production of the vegetation type distribution map of Qinghai Lake Basin was completed.Compared with the 1:1,000,000 vegetation map,the vegetation type map of the Qinghai Lake Basin has three new types of vegetation that were previously neglected due to too small or mixed growth: Elymus,Bupleurum and Pedicularis.The number of plaques increased from 190 to 13,690,adjusting the boundaries of each classification unit,refining the distribution area of each classification unit,and better reflecting the detailed characteristics of the spatial distribution of vegetation in the Qinghai Lake basin.