Vegetation Type Recognition Based On HJ-1A Hyperspectral Imagery

Vegetation has obvious interannual variation characteristics and is considered as an important and sensitive index to reflect the change of ecological environment.Remote sensing technology has small time interval and wide coverage area,which is beneficial to the identification of vegetation types and monitoring of coverage changes.Hyperspectral imagery which is rich in spectral information and spatial information fusion characteristics of image distribution can help to distinguish vegetation types.The HSI data of HJ-1A satellite is the first domestic space hyperspectral image data in our country,but the time in orbit is short,and the case of it is still few,in addition,the vegetation coverage of Golmud in Qinghai province is low,the soil type is mainly saline-alkali and the ecological environment is very fragile.It is important to study the identification and distribution of vegetation types in this area for the application of HSI data and the ecological environment of the area.This paper is based on a review of studies on hyperspectral remote sensing at home and abroad and related classification after dimensionality reduction,the image data from 2010 to 2015 of HJ-1A satellite is selected.According to the characteristics of HSI data,the preprocessing of image data is completed,and the research areas is divided into the northern and western regions.Then,using isometric mapping(Isomap)dimensionality reduction algorithm,which can keep the global topological characteristic of the image and the distance between sample points and retain the valid information,to extract the dimension feature of HSI data,and reduce the number of correlated bands.Based on the field investigation data,selecting sample data of training areas and validation samples,and combined with support vector machine(SVM)classification method for vegetation type recognition and analysis and evaluation of precision,and analysis of vegetation distribution characteristics and temporal vegetation change characteristics.The method of vegetation type recognition based on Isomap and SVM shows that the method has higher precision in the identification of vegetation types in low vegetation coverage area,and it has certain feasibility.The overall classification accuracy in the north and west of the study area is 80.86% and 80.26% respectively.Through multi-temporal area change analysis,it was shown that the area of each vegetation type in the two regions did not change significantly within 5 years,and basically remained within 5%.Combined with DEM and classification results,the boundary of vegetation distribution in the two research areas is consistent with the DEM grade line,and the topographic changes are the influencing factors of the vegetation distribution in the two research areas.The distribution of vegetation in the northern part of the study area takes the height of 2723 m and 2758 m as the boundary of vegetation type;the distribution of vegetation in the western part of the study area takes the height of 2786 m and 2815 m as the boundary of vegetation type.