Fully Constrained Estimation Abundance With Primal-Dual Interior-Point Method

Abundance estimation has been one of the important techniques for mixed pixel in hyperspectral image. The so-called abundance estimates, were estimated endmember of the proportion in mixed pixels. Abundance estimation technology needs to meet two constraints based on the actual physical meaning, abundance non-negative constraint (ANC), and abundance sum-to-one constraint (ASC), satisfied constraints of ASC and ANC abundance estimates technology is fully constrained abundance estimation techniques.Based on the linear spectral mixture model, through the analysis of existing N-FCLS, EES-FCLS, IIM-FCLS, PDIP algorithm, this paper mainly completed the following two tasks. First, the full constrained abundance estimation algorithm of linear spectral mixture model is less noisy, and it is difficult to improve the accuracy of abundance estimation. In this paper, the error caused by the error in the form of noise is expressed in the form of noise, that the noise size is called the sum-to-one deviation p. According to the actual physical meaning, due to the influence of ground environment complexity, noise, light, endmember purity etc. many factors, resulting in p and spectral root mean square error are strongly related to each other. And only find the best accuracy of mixed pixels, the fully constraint abundance estimation can achieve the best effect.Based on the above conclusions, this paper combined with the primal dual interior point method, improved the constraint abundance estimation algorithm, culminating in the optimization FCLS+PDIP (including ρ) algorithm and optimization FCLS+PDIP (N=1,ρ) algorithm. The optimization FCLS+PDIP (including p) algorithm also inherited the characteristics of IIM-FCLS and PDIP algorithm, and then the reconstruction error as a end conditions added into the algorithm, making the algorithm of p adaptive, so as to improve the accuracy of the algorithm. Due to the improvement of the accuracy of the algorithm is mainly concentrated in the first iteration, in order to take into account the running time of the algorithm, can limit the number of iterations, so N=1, through the sacrifice of partial decomposition accuracy, to speed up the decomposition of time. Through the simulation of hyperspectral data, real-world hydice data and real hyperspectral spilled oil data, validate the algorithm in decomposition has greatly improved the accuracy.At the same time, it also has a high efficiency in decomposition time. So the algorithm is a more efficient algorithm.