Based On Multi-source Remote Sensing Data Of Poyang Lake Wetland Vegetation Types Inversion Study

Remote sensing technology is widely used in the wetland vegetation monitoring field, such as wetland vegetation identification and classification, lots of research to achieve better results is based on the multi-spectral remote sensing images. But the traditional multi-spectrum remote sensing data has spectral resolution is usually decades to hundreds of nanometer, however, to reflect material differences in spectral characteristics of the absorption peak or the reflection peak width is generally around5-50nm, therefore the fine material classification will need the higher spectral resolution. High spectral characteristics of the researchers on wetland plants is not yet sufficient ground studies, is not a comprehensive wetland plants spectrum library can be used to identify remote sensing identification. Moreover, a single remote-sensing data can’t accurately to reflect the surface features. High spatial resolution remote sensing image has high spatial resolution, but it contains less of spectrum information. Hyperspectral remote sensing image has high spectral resolution, but its low spatial resolution, and is more susceptible to the effects of the atmosphere, which cause it can’t accurately identify the features of the area. Using multi-source remote sensing data and its fusion image, it can realize the information complementary, which can extract more information and improve interpret precision.This study used for a variety of remote sensing data about Poyang Lake in support of the "3S" technology, and combine it with the analyzed ground-research data. In order to survey and research the wetland vegetation resources distribution types, size, etc., so as to provide background information for the distribution of wetland. Moreover, it can be used to management multi-level information of the Poyang Lake wetland, then to effectively protect wetland eco-environment and wetland vegetation remote sensing classification and mapping.First, based on analysis of collected ground vegetation spectral reflectance data and establishment of wetland vegetation spectral database, we select the appropriate hyperspectral remote sensing data and high spatial resolution remote sensing data.On the remote sensing data was preprocessed, in accordance with the region of interest area cut for study area, hyperspectral image and the hyperspectral-high spatial resolution fusion image for vegetation classification. The vegetation classification was using SAM, ISODATA, and SAM+ISODATA classification methods to classify; the fusion image analysis to come up with the optimal fusion method. Moreover, the best classification method can be used for the image by the fusion method processed. According to the research, draw the following results.(1) The research7plants spectrum characteristics of the spectrum with vegetation characteristics, and relatively easy to distinguish, however, in the part of the wavelength range P. arundinacea L. and T. Lutarioriparia L. Liu spectral characteristic curve of confusion. The" second derivative characteristics of the vegetation spectrum in685.8nm,692.4nm,698.6nm,704.3nm,737.5nm,746.2nm,957. lnm seven best band Recognition vegetation.(2) The accuracy of SAM classification of the hyperspectral image classification was slightly higher than the ISODATA classification. SAM classification can identify four types of vegetation, and ISODATA classification method can only identify three kinds. Interpret the vegetation type area, the accuracy of various classification schemes to identify the vegetation is slightly different, generally classification accuracy of P. arundinacea L. was highest.(3) The hyperspectral and high resolution image fusion using three fusion methods, such as wavelet transform and Gram-Schmidt transform and Principal Components Transform. The wavelet transforms fusion has the best quality, and Gram-Schmidt fusion method minimum. For singly band, principal component analysis transform for the53band of the image fusion of the best quality, while the wavelet transform fusion of the best quality for the first85-band and108band.(4) Based on analysis of the hyperspectral and high spatial resolution image transformed fusion images, wetland vegetation classification accuracy and identification of vegetation types were higher than those singly hyperspectral remote sensing images. The fusion images vegetation classification accuracy is86.75%and there were6species identified, while the hyperspectral image classification accuracy is79.12%and only identified4species with vegetation.