Study On Wetland Information Extraction Using Multi-angle Hyperspectral Chris Image Data

Wetlands are a unique ecosystem in earth,containing the most biodiversity and complexity ecological landscape.Accurately acquiring the vegetation information from different types of Wetlands,which is the important aspect in wetland evaluation and wetland monitoring,Can provide effective technical support for the Wetland protection and restoration.Spectrum of wetland vegetation is not only highly similarity,but also has the strong spatial variability.There are different spectrums of vegetation from the different viewing angles.Hyper-spectral remote sensing technology Can identify small differences in spectral changes.The unique properties of multi—angle hyper-spectral remote sensing Can be use for the vegetation information extraction to improve the identification and classification accuracy in wetland.In view of this,a comprehensive and systematic study conducted on vegetation information extraction in Long Baotan wetland from CHRIS with five angle images.We use the two classification methods,SVM and SAM,to inverse the wetland types and vegetation information from the different angle images and their fusion images.By comparing the classification results,the best band combined mode and image fusion mode of different angle images were determined.The main results and conclusions are as follows:(1)The angle effects of vegetation indices,which can be used in fusion image,is benefit to improve the accuracy of wetland vegetation information extraction.The paper provides a more effective fusion mode between the vegetation index and different angles image,fusioning the NDVI of一36。image with 0°image.The result showed that the extraction accuracy of the wetland vegetation information approach to 92.23%,while it was only 67.04%ifthe SAM was used directly to the 0°image.The result also indicated that multi—angle and hyper-spectral remotely sensing data have important application potentiality in extraction of wetland vegetation information. (2)It is possible that the outline of flood wetland Can be draw based on the distribution of dominant species of wetland vegetation extracted from the hyper-spectral image,in according to the dominant species of wetland vegetation does not change both in the flooding period or in the dry season.The paper achieves beuer result on drawing the outline of flood wetland in Long Baotan area.(3)There exists the angle effect in the spectral reflectance.The transform of different angles image and band combination can be used to wetland classification.The studies showed that a new color composite image of RGB include more surface information,by combining with 57-band(940nm)of+36°image,0°humidity image and 4-band(461nm)of -36°image.The classification accuracy by SVM on the new image is 92.52%,which was greatly improved then 76.1%of traditional supervised classification accuracy.It also provides an effective method to extract wetlands information.