Study On Classification Of Poyang Lake Wetland Area Based On Landsat TM Image

Poyang Lake is the largest fresh water lake in China and a renowned wetland in the world. The state council of China approved it to be state nature protection area in May 1988, and rated to the important swamp scroll of world "swamp convention" by UNESCO in July 1992.This paper was primarily conducted to probe into the method and approach which could improve classification accuracy of wetland, made experiment of the computer automatic classification methods to Landsat TM Image covering Poyang Lake wetland. In experiment firstly used unsupervised classification to remote sensing image, made progress of manual work interpreted to make certain their attributes, then transformed their outcome to the file of classification cyclostyle, and then carried out supervised classification, it indicated that the classification precision of Poyang Lake wetland was improved through these means.In supervised classification,this method was illustrated comparatively with classification carried on TM image of three different wave band synthesizing, multispectral synthesizing image, principal component analysis synthesizing image after Karhunen—Loeve transform. We could have the related conclusions as follow:At first, the classification overall accuracy of integrating unsupervised and supervised classification procedures applied to the images of TM4, 3, 2, TM3, 4, 5 , TM7, 5, 3 ,principal component analysis and multispectral synthesizing was 82.53%, 84.12%, 86.85%, 88.20%, 91.73% respectively.Secondly, wave band 5 containing abundant spectral information, it is the best wave band in wetland remote sensing classification. It was attributed to that water, which is the important factor working on distributing of wetlands, is mostly sensitive to the wave band 5 and band 7. It was mostly redounded to distill vegetation synthesizing red wave band and near infrared. Furthermore, it was beneficial to enhance the Poyang Lake wetland classification using TM7, 5, 3 image with remote sensing technique.Thirdly, the each wave band of the image after principal component analysis could be largerly contributed information to new image, made the new image to be interpreted more easily, could be improved the efficiency of data processing in the premise of assuring the precision. Multispectral image has more information, then has the best result of interpreting, but containing more redundance data.Fourthly, the diference of distributing stratified random sampling has affected the accuracy of wetland classification to a certain extent when assessing the classification result with error matrix.Last but not the least, presently interactive interpretation approach had to be applied in the study for the sake of higher classification accuracy of wetland with remote sensing technique. From the practical point of view, the diference of interpreters, the knowledge of related bio-geography, the proficiency of mastering the ERDAS IMAGINE software, the familiarity with the research area and so on, has a remarkable influence on the classification accuracy. The cloud cover was lower and the image quality of research area in the Landsat TM Image, selected in the study, which was one reason of the higher classification accuracy.