Research And Implementations Of Water Extraction Of Urban Water Network Based On Hyperspectral Remote Sensing

Hyperspectral data has the characteristic of numerous and continuous spectral bands, combining image information with spectral information. It can provide a wealth of detailed surface information and capture the complex characteristics of inland open water bodies. Based on characteristics of hyperspectral remote sensing data, analysis and preprocessing are conducted in this thesis. Some typical water extraction methods are implemented and accuracy assessment and performance analysis are made.The main works are as follows:(1) We analyze the theoretical spectral model of the city’s land cover, including: water, vegetation, soil, urban and shadow, and made a summary about spectral reflectance characteristics of each feature. Detailed analysis on the characteristics of hyperspectral data sources and pre-work are made, the data models are also described. We invent a new method on quick view generation of Remote Sensing Image based on Linux, making it more convenient for analyzing remote sensing image.(2) The rivers are crisscrossing in the southern river city, and there are many factors like shadows and other features which affect the accuracy of water body extraction. By analyzing different factors affecting results of water body extraction, we design five scenes including different water body environment from the AISA airborne hyperspectral images. Then we find water characteristics band:690nm-770nm, despite the near-infrared bands. It provides more information about water body which is different from other land covers.(3) The existing water extraction methods are analyzed, including:Spectral structure method, Normalized Difference Water Index method, Support Vector Machines. Based on the characteristics of hyperspectral remote sensing data, we select different characteristic bands applied to the classical algorithms to examine the extraction of water bodies from different scenarios. According to the feature bands achieved from scenario analysis, we construct a new tree, which combined near-infrared threshold judgment and Water Index images. We apply this new method and compared the results to the existing water extraction methods, then we find that the feature-based decision tree algorithm is a more convenient, fast and accurate method in urban water extracting.(4) The comparison between the results of Spectral structure method, NDWI, SVM and the decision tree based on the water spectral characteristics, shows that the feature bands about water from hyperspectral data sources have a perfect performance on suppressing non-water information, especially in the shadow of urban architecture. This discovery lays the foundation for future urban terrain classification, particularly for the extraction of water research.