Vegetation Classification Based On The HSI Hyperspectral Data Of HJ-1A Satellite

Hyperspectral Remote Sensing technique is the leading edge technology in the field of remote sensing, which provides a new technical means for the acquisition of information of plant species or populations and the recognition of a wide range of vegetation. In the classification of vegetation, because some types of plant communities have similar spectral characteristics, the application of multi-spectral and other traditional remote sensing images will be greatly affected. Hyperspectral Remote Sensing has extremely high spectral resolution, so the accuracy of recognition and classification of vegetation will be highly enhanced. We can screen out the bands which have obvious spectral differences in vegetation types from a large number of narrow bands, and use a few narrow bands to identify and classify vegetation types. Or we may regroup several integrated bands with compression technology, so as to make full use of the spectral information of vegetation and improve the accuracy of identification and classification of vegetation.Launched in2008, the small satellite constellation of environment and disaster monitoring and forecasting (HJ-1A) is China’s first satellite special for environment monitoring. The contiguous band settings of HSI hyperspectral data acquired by the HJ-1A Star can reflect the subtle changes of surface feature spectrum and enhance the accuracy of recognition and classification of the ground features, therefore, its application can meet our country’s needs of dynamic monitoring of vegetation on a large scale. However, due to the limitation of a short launch time, currently there are just relatively few vegetation classifications and related researches based on HSI data. This paper takes Jinzhongshan Natural Reserve as the research area, utilizes hyperspectral HSI data provided by the small environment satellite, and focuses on the following researches:1. Introduction to and pretreatment of HSI data:introduces the acquisition, characteristics, and contained spectral information of HSI data. Aiming at the characteristics of HSI data products, pretreatment work includes data format conversion, removal of poor quality bands, and repairing bad lines on several bands for HSI data, removal of vertical stripes existed in HSI data and finally the geometric precise correction. 2. The Spectral of Difference vegetation types:We determined7types of ground features in accordance with certain principles, selected training samples in the image combining with the1:200,000vegetation distribution map, rare vegetation distribution map and field actual measurement data and other data of the research area, and picked up the average spectral curve of each ground feature to analyze their spectral features.3. Comparative study of different classification methods:Based on a deep analysis of differences in spectral characteristics of each ground feature, we researched on the classification of vegetation in the natural reserve, adopting commonly used the decision tree model, supervised classification methods including the maximum likelihood classification, ISODATA classification which belongs to unsupervised classification, and spectral angle mapping of spectral matching method. By contrasting the classification accuracy of different methods, we found that the decision tree model and likelihood classification had the highest accuracy, reached83.74%,76.22%。 This results indicating that this method for vegetation classification can effectively classify and identify vegetation with high accuracy, and has certain universality for classification of vegetation in different regions with the same data resources.After the MNF transform of HSI hyperspectral data we also reached a high accuracy based on supervised classificationGenerally speaking, by referring to HSI hyperspectral data this research fully analyses the spectral characteristics of typical ground features in the research area, and by classifying vegetation in the research area with different methods, it sufficiently excavates potentiality of HSI hyperspectral data for vegetation classification and successfully expands the application field of remote sensing with HSI hyperspectral data applied to vegetation classification and identification. It not only confirms the feasibility of HSI data of HJ-1A star applied to vegetation identification and classification, but also plays an important role in promoting the application of China’s satellites and remote sensing technology to relevant areas.