Remote Sensing Monitoring Of Yutian Reservoir Based On Unmixing

Yutian reservoir serves as the backup water source of jingdezhen city,while the traditional reservoir water information automation monitoring accuracy is poor,the measurement results are inaccurate.In order to better manage the Yutian,so it is necessary to monitor the water change information of Yutian reservoir.This paper,we select"Yutian lake"as the research object,with monitoring the water area and the water change detection as application background,using Landsat8satellite data,apply mixed pixels unmixing method and multi-source information combinatorial technology to automatic remote sensing monitoring of Yutian reservoir.The paper focuses on how to use structured sparse unmixing technology to improve the accuracy of the water area calculation.The research involved as follows:An approximate sparse constraint of multilayer nonnegative matrix factorization?AL₀-MLNMF?hyperspectral algorithm has been proposed,this algorithm is improved on the approximate sparse constrained non-negative matrix factorization model.the observation matrix has been multi-level decomposition,decomposed into the multiplication form of several small non-negative sparse matrix.this method effectively improved the precision of unmixing.And study the abundance coefficients of structured sparse model and analyze the mathematical properties of structured sparse regularities.Then,established a high model,and analyze the convergence and the condition of the model,study the numerical solving method,and analyze the influence of each parameter on the unmixing performance.And comparing and evaluating several common hyperspectral unmixing algorithms.The experimental results show that this algorithm unmixing accuracy is better than others.An approximate sparse constraint space total variation multilayer non-negative matrix factorization?AL₀TV-MLNMF?hyperspectral algorithm has been proposed,this algorithm is improved on the AL0TV-MLNMF model,and considering the complex spatial geometry of hyperspectral remote sensing image,which increases the spatial total variation space model.Compared with several common sparse algorithms,the results show that the accuracy of the algorithm is greatly increased.A method of water change detection based on mixed image unmixing and multi-source information fusion is proposed.Traditional water change detection methods have many limitations,so it is very difficult to detect the change of water for long time.First study water area change detection method based on hyperspectral remote sensing technology,aiming at the problem of hyperspectral remote sensing image existence mixed pixels,analyze the hyperspectral imaging spectral information and spatial structure information,trying to build a more suitable to the actual unmixing for structure sparse spectral model and high order non-isotropic total variation spatial structure model,and introduced the AL0-MLNMF model and AL₀TV-MLNMF model into the waters change hyperspectral remote sensing image unmixing,then use the water body parts of unmixing result in the water area change detection.Based on the test and analysis of the10 satellite data obtained by Landsat8 satellite from 2013 to 2016,the validity of the algorithm was verified.