Reservoir Surface Water Extraction And Monitoring Based On Pixel Unmixing

Recently, with the development of remote sensing technology, remote sensing technology has been widespread concerned, and was widely used in different area. Among them, the use of remote sensing technology to extract water and dynamic monitoring technology can be widely used to assess floods, water environment monitoring, water resources testing and other research areas. Landsat images have the advantage of run short cycle, data volume, a large number spectral bands and download free etc. For water remote sensing applications require a higher monitoring frequency, is a suitable data source.Existing water extraction methods mostly based on pixels, but the presence of mixed pixels limit the accuracy of water extraction. So deep into the pixel interior and unmixing pixel, extracting water from the perspective of sub-pixels, will improve the accuracy of water remote sensing extraction.The research is focused on water mixed pixel based on unmixing model, water extraction model, and water spectral feature. For the problem of land-water borders mixed pixel, shadows, and dark surface, using Landsat images of Xinglin Bay reservoir and Shidou reservoir to experimental verification. The main researches are as follows:Firstly, briefly introduced the natural situation of the study area and experimental data, and data preprocessing operation; Learn water and other surface features spectral characteristics, studied water extraction model, and using water index and SVM classification to extract water and analysis results.Secondly, based on the theory of mixed pixels, understanding existing unmixing model and unmixing process; select fully constrained least squares method(FCLS) unmixing method to extract water, and use Google Earth TM ground truth data with high spatial resolution as a reference, to evaluate the accuracy of the extraction results. The results showed that the FCLS extracted result is better than water index method and SVM classification.Finally, compute Grey-Level Co-occurrence Matrix(GLCM) of study area, improve the existing FCLS method by fusing texture information. The accuracy of extract results was evaluated based on Google Earth TM with ground truth data. By comparison of the two water extraction results and show that improved FCLS method has higher accuracy. Combine FCLS method and PCA to prose FCLS-PCs method for reservoir surface water change detection, mostly in detecting the changes between two different times. Comparison the detection method and the results show that the accuracy of the detection method is relatively high. Using FCLS method and multi-temporal remote sensing data to dynamic monitoring reservoir surface area, and analyze the reasons of change; it is contributed to manage the ecological environment of reservoir.