Research On The Method Of Water Body Extraction Based On Remote Sensing Image

With the continuous development of remote sensing technology, its application field has been widened. Using remote sensing images and other ancillary data for extraction of water information is an important means. In this thesis, on the basis of several existing kinds of water extraction model and through experiment verification, the thesis determined a suitable method for the extraction of plateau lakes so that the shadow and the influence of mixed pixels could be eliminated.This thesis mainly was based on the remote sensing images of TM/Landsat 8 OLI of the Qinghai Lake in 5 periods and used the band operation classification method, Supervised Classification, Decision Tree Classification and relations between single band and multi-band spectral combination method to extract Qinghai Lake in nearly two decades. By using artificial visual interpretation and vector method the thesis verified the precsion of extraction methods above. Draw the following conclusion:(1) The results using single-band threshold method or multi-band spectral relations were compared with the results of the vector quantization artificial, and the result is not very ideal Both of them greatly improved the accuracy of both methods. And Modified Normalized Difference Water Index (MNDWI) method for the study area had a good effects, it reveals the microscopic features of the water to a great extent so it was suitable for extraction of the plateau lakes.(2) By using the data that enhanced by K-T transform and Normalized Difference Water Index (NDWI), based on features spectrum, this article extracts water information of Qinghai Lake with decision tree classification. The result represents that the overall precision reaches 91.26%, and Kappa coefficient is 0.8381. Compared with artificial classification result, the relative errors is the most close, which means that decision tree classification is the most effective means to extract the plateau lake water information.(3) Calculating the water area of Qinghai Lake with different water information extraction models, the result shows that the decision tree classification fits the study area best when compared with artificial classification result. Using decision tree classification, calculating the water area of Qinghai Lake during from the year 1995 to 2004 with the remote sensing images, analyzing the dynamic change, we can know that the area of Qinghai Lake has the trend of decreasing first and then increasing, and the Qinghai Lake shrinking first and expanding then. Explain the main reason of this situation and propose a suitable control measures to Qinghai Lake area.