Mixed Pixel Unmixing And Sub-pixel Mapping On Hyperspectral Imagery

With high-resolution remote sensing satellites are launched, massive remote sensingimage is obtained, but we lack real-time analysis and processing ability, one of the mostdeficient are all kinds of efficient algorithms. For hyperspectral remote sensing image, mixedpixel decomposition is the most urgent to solve the problem, which directly restricts thepractical application of the image. But it is not enough to solve the problem of mixed pixels,it can only get the endmember abundance, and can not determine spatial locationaldistribution of the sub-pixel, so we need to solve the problem of the sub-pixel location. Atthis point, hyperspectral image can universally be applied. This paper describes the basic concepts of hyperspectral remote sensing, studies thecharacteristics of hyperspectral image, summarizes the existing techniques of mixed pixeldecomposition, and focuses on the analysis of several common endmember extractionalgorithms. It also summarizes the existing sub-pixel locational technologies, and uses theprogram to achieve a sub-pixel locational algorithm. Finally, by summarizing the existingtechniques of hyperspectral mixed pixel decomposition, puts forward the optimizationalgorithm based on vertex component analysis.Vertex component analysis (VCA), its essence is a kind of pure mathematics method,which has a good theoretical foundation, achieves good effect. But VCA algorithm has threedefects: Without considering image spatial information, its effectiveness may reduce forlarger noise hyperspectral image. Algorithm requires to predetermine the numbe ofendmembers, but it is very difficult to determine correct the number in advance. Thealgorithm runs in many times, but results are unstable. To solve above problems, this paperproposes the improved VCA algorithm (Improve-VCA), specifies the number of candidateendmembers, uses the iterative calculation of candidate endmember interval, combines withthe image spatial information and optimization mechanism of ill-conditioned matrix tocircumvent judgment. Finally, the VCA algorithm is improved.In order to quantitatively evaluate the algorithms, and fully prove the correctness andeffectiveness of thought in practice, this paper generates simulated hyperspectral data,contrasting the simulated data for the common endmember extraction algorithms (N-FINDR,SGA, VCA and ACEEHIIU) and the proposed algorithm (Improve-VCA) in the experiments and tests them in the same conditions, and makes the strictly quantitative analysis anddescription. The indicators of quantitative study use the average spectral angle mSAD, theaverage spectral information divergence mSID, component average angle mAAD and thecomponents overall root mean square error of inversion of the abundance mARMSE to makethe comprehensive evaluation and analysis. By contrast, analysis shows that this algorithmcan automatically determine the number of endmembers, accurately extract the endmembers,can be comparable with common endmember extraction algorithms in many respects, even insome respects, this algorithm is superior to the common endmember extraction algorithms.Finally, this paper uses code to implement a sub-pixel mapping algorithm based on MAPmodel, and carries on the preliminary study about the sub-pixel location, lays the foundationfor future research.