Research On Analysis Of Hyperspectral Characteristics And Extracting Characteristic Bands Of Typical Wetland Vegetation In Poyang Lake

Cause the limitation of spectral resolution and band number,it is hard to obtain the local fine information,reflect the spetrum information and analyze the differences of typical vegetation by traditional multispectral data.There are some difficulties in vegetation classification and recognition using multispectral data.In contrast,the characteristics of hyperspectral remote sensing data are more number of channels,higher spectral resolution,narrower bandwidth,It has great potential to analyse the difference of reflectance spectrum of wetland vegetation.According to the characteristics of similar spetrum and high difficult classification for different wetland vegetation,the hyperspectral data of six kinds of typical vegetation in Poyang Lake wetland are measured by spectrometer,those typical vegetation included Cynodondactylon(Linn.)Pers,Triarrhena lutarioriparia L Liu,Polygonum hydropiper,ArtemisiaselengensisTurcz,Carex cinerascens Kükenth and Phragmites australis.Base on the pretreatment of original spectral data,the research use a series of methods to transform the hyperspectral data and analyze the spectral characteristics of different vegetation,those methods including derivative,Log(1/R),continuum removing and so on.Secondly,one new method is comed up with to extract the spectral characteristic bands,which base on data error range and mean difference.With the existing method,including the the method of Mahalanobis Distance and the method of Mean standard deviation of spectral,three method are apply to various mathematical transform to select their spectral characteristic bands and analyse the effect of three method,using T-test to detect the sensitivity of characteristic bands.Based on the spectral characteristic bands,six kinds of typical wetland vegetation were classified by Bayes and Fisher Discriminant Analysis and the accuracy results are evaluated.Finally,we use Landsat8 images to examine the differences between different vegetation levels.The hyperspectral data has a good representation of the spectral characteristics of vegetation information,six kinds of typical vegetation has similar spectral characteristic curves.Through mathematical transform correlation statistics,various spectral parameters can effectively highlight the spectral differences among different vegetation,and it is conducive to the identification of different vegetation types.After the envelope transformation,the difference of the six vegetation spectra in the four bands was the most obvious,those bands were 400-530,550-730,930-1000 and 1100-1250 nm.Compared with the existing spectral feature extraction methods,the spectral band selection method proposed in this paper is better.The characteristic bands of original spectra are 422-452 nm and 530-559nm;the characteristic bands of Log(1/R)are 422-452 nm and 522-597nm;the characteristic bands of CR are 400-529 nm and 558-567 nm and 653-670 nm and 1173-1184 nm and 1209-1239 nm.Based on the spectral characteristics of different bands,which can effectively discriminate vegetation spectral classification using Bayes discriminant analysis and Fisher discriminant analysis,and classification accuracy is higher than that of the original through spectrum transform spectrum,some vegetation envelope transformed by the precision can be improved by 13%.Using Landsat 8 images can distinguish the partial vegetation effectively,but due to the mixed pixel,part of the vegetation is not indistinguishable from the component.The research results lay a foundation for the wetland remote sensing information interpretation and wetland vegetation mapping,also provide a scientific basis for the protection of the lake wetland vegetation and the ecological environment of lakes.In addition,it also has reference meaning to the Spectral Recognition and classification for the freshwater lake wetland vegetation.