Classification And Degradation Index Extraction Of Grassland Vegetation Based On Hyperspectral Remote Sensing

Grassland is an important natural resource in the three-river headwaters region.In order to achieve efficient and accurate resource utilization and protection,the application and development of remote sensing technology is an urgent problem to be solved.Hyperspectral measurement technology enables satellite images to contain more information,which makes it possible to extract refined grassland information.Taking Xinghai county as an example,the use of domestic resources.ZY1 E satellite observations of hyperspectral image,combined with field investigation,the grassland vegetation types in the study area and its spectrum characteristic is analyzed,and the use of a variety of feature recognition method and mixed spectral decomposition method separately like yuan scale field like yuan scale and the information extraction,Furthermore,the ideal method of using hyperspectral remote sensing to carry out grassland vegetation classification mapping and degradation index extraction was explored.The main research conclusions of this thesis are as follows:(1)In the analysis of spectral characteristics of typical grassland vegetation,low coverage grassland can be identified by the reflectance of chlorophyll sensitive band(559nm and 670nm)in visible band,while in the near infrared band,the spectral curve of each vegetation fluctuates greatly,and the reflectance characteristics at 1122 nm and1660nm,It can distinguish shady slope vegetation,Kobresia capillifolia grassland community and crops from other vegetation,and the first derivative transformation and envelope removal transformation can enhance the spectral characteristics of each vegetation type in "red edge" and "red valley",respectively.(2)Combined with Mahalanobis distance method and correlation analysis,the reflection spectrum optimization and redundancy were carried out for 12 dominant grass species,such as Kobresia pygmaea,Kobresia humilis and Stellera chamaejasme,and eight sensitive bands,507 nm,670nm,713 nm,765nm,1257 nm,1324nm,1459 nm and1929nm,were obtained.(3)Among the three commonly used hyperspectral remote sensing classification methods,spectral Angle mapping,spectral information divergence and decision tree,the decision tree method based on spectral feature analysis results had the highest overall classification accuracy(77.56%),and achieved a better effect on the grassland community level.(4)Multiple Endmember Spectral Mixture Analysis(MESMA)was used to study the Spectral unmixing of typical grassland features.The results showed that MESMA showed good spectral decomposition accuracy between vegetation and bare soil(RMSE0.16),but poor spectral decomposition accuracy between fine pasture,weed and bare soil(RMSE 0.28).(5)Two remote sensing indicators,"vegetation ratio" and "fine pasture ratio",were established based on spectral dismixing results of hyperspectral images to characterize grassland degradation,and grassland degradation was graded in the image range.The results showed that the grassland in the image was mainly moderately degraded(56.25%).