Hyperspectral Data Decomposition Of Mixed Pixels, Validation And Spectral Matching Algorithm

Hyperspectral Remote Sensing (HRS) is an important way for quantitative analysis of remote sensing. It has a broad application prospects. But there are a lot of mixed pixels in the hyperspectral data, this situation becomes an important bottleneck problem that affect the accuracy of the remote sensing applications. Therefore, in order to improve the accuracy of HRS applications and make HRS applications achieve subpixel precision, we must solve the problem of mixed pixels decomposition. We systematically research several unmixing algorithms and put forward some improvements in the direction of mixed pixel decomposition for hyperspectral data. Then we use spectral matching to verify the unmixing algorithms.Based on the reviews of the research status about hyperspectral mixed pixel decomposition and spectral matching, the main research works and corresponding creations are listed as follows:1. We research the endmember extraction problems based on the linear spectral mixture model. And then we achieve several classic endmember extraction algorithms. We design and achieve four endmember extraction algorithms including PPI、VCA、N-FINDR and ATGP. On the basis of those researches, and then we put forward two improved algorithms in allusion to N-FINDR. The experimental result shows that the improved N-FINDR algorithms accelerate the convergence of the algorithm and improve the accuracy of mixed pixel decomposition.2. We research the abundance estimation problems based on the linear spectral mixture model. And then we achieve four abundance estimation algorithms on the basis of the method of least squares. The experiments give the performance analysis and experimental verification for four algorithms by different test sets.3. Firstly, the paper researches the problem of the sparse linear mixed pixel decomposition. And then we design a mixed pixel decomposition algorithm using the method of alternating iteration by establishing a cost function for the sparse mixed pixel decomposition problem based on norm constraint. The result shows that this method can get better abundance results than unconstrained least squares method. And we can use this method to find endmember and abundance without assuming pure pixels existing.4. On the basis of the classic spectral matching algorithm introduce including spectral encoding matching, spectral angel matching and spectral correlation matching, we put- forward a new matching algorithm by coring the information of the spectral curve peaks and valleys. We take advantage of the numbers of the spectral curve peaks and valleys in the local area, and then we use grouped encoding to get the shape matching of the spectral curve. The experimental results show that the method has higher spectral matching accuracy and verify its performance in object identification by combining with five endmember extraction methods.