| Eupatorium adenophorum Spreng., a worldwide noxious weed, invaded to China from theboundaries of Vietnam and Burma in 1940s. It is especially rampant in west-south of Chinaand have a tendency to invade to the north continuously. The traditional investigation is tooexpensive and low accuracy to carry out monitoring measures timely. In this paper, Xichangcity in Sichuan province was taken as the study area and classification of the land cover typebased on images of ASTER received on March 3 of 2005 and land-use map were conducted.Firstly, through the investigation of the dominant land cover types in the study area, theprinciples used for the classification were determined based on the practical conditions of thestudy area by using remote sensing images, and the samples' indices data of Eupatoriumadenophorum Spreng community were got meanwhile. Secondly, the optimal bands whichwere chosen by statistical analysis of ASTER raw datasets were transformed to fusion image,and texture information of image were derived from the most appropriate vegetation indexwhich was selected by conducting and correlation analyzing community indices andvegetation indices such as NDVI,RVI and PVI. Thirdly, the six different combinations ofthe land cover type classification were tested by the different classification methods such asthe maximum likelihood classifier and Mahalanobis distance classifier. By the comparison ofsix types of classifications and validation of ground truth data, the mostly suitable one withthe highest accuracy to the land cover type classification in the study area was determined,which would apply to monitor distribution of Eupatorium adenophorum Spreng effectivelyand timely.Through this study the follow conclusions are drawn:(1)Through conducting and correlation analyzing of community index and vegetation indexin research area, the results show RVI has the highest sensitivity to the community index among NDVI,RVI,PVI, which correlation coefficients are above 0.7, and reflects communitycharacteristic sufficiently.(2) Gram-Schmidt fusion is tested, that improved spatial resolution from 30m×30m to30m×30m, and spectra information are not changed in fusion meanwhile, which ensure thereliability of following classifications.(3) To the same image, the study has shown that maximum likelihood classifier is better thanMahalanobis distance classifier, for the overall accuracy and Kappa coefficient of themaximum likelihood are 78.50% and 0.7496 respectively, greater than the latter, which are73.70% and 0.6936. Furthermore, the overall accuracy and Kappa coefficient of themaximum likelihood will arise to 90.52% and 0.8894 respectively if adding the spatialtexture information on the fusion images, better than the latter, which are 80.64% and0.7741.(4) To the same classified method, the study has also shown that the result of the fusionimages adding the spatial texture information derived from RVI by the maximum likelihoodclassifier is better than that of without texture information, for the former' overall accuracyand Kappa coefficient are 90.5207% and 0.8894 respectively, which are better than bothfusion images without the spatial texture information and VNIR image. To the Mahalanobisdistance classifier, the result of the fusion images adding the spatial texture information isbest, which overall accuracy and Kappa coefficient are 86.2483 % and 0.8397 respectively,and the result of the VNIR image is less than that those are 80.6409% and 0.7741respectively, and the result of fusion images without the spatial texture information is theworst that thoseare73.6983 % and 0.6936 respectively.(5) In the accuracy of simple land cover type, accuracy of types that are related to vegetationsuch as Eupatorium adenophorum Spreng,grassland and forest rise by adding textureinformation, which are 86.92%,73.91% and 100% respectively by using the maximumlikelihood classification.(6)To compared the results of six classification methods, the study has shown that the bestway is by using the fused images of 2,3N,4bands which adding the spatial textureinformation derived from RVI, and using the maximum likelihood classification that theoverall accuracy and Kappa coefficient of the classification is 90.5207% and 0.8894 respectively,. This accuracy can basically meet the demands of the classification and mappingof remote sensing for Eupatorium adenophorum Spreng in the study area.
|
PDF Full Text Request