Nonlinear unmixing of hyperspectral images based on double-bird flock optimization

Aiming at the characteristics of nonlinear mixing on end-member materials in hyperspectral image pixels, a nonlinear unmixing algorithm of hyperspectral images based on double-bird flock optimization was proposed referring to the phenomenon of biological swarm intelligence. In order to further improve the accuracy of nonlinear unmixing algorithm, the optimal solution of the nonlinear problem was obtained by simulating the behaviors of foraging, vigilance and flight in the bird flock. The algorithm alternately updated the optimal solution of the objective function and the nonlinear model parameters through iteration optimization of double-bird flocks, and finally obtained the optimal estimation of the end-member abundance on hyperspectral images. The experimental results of simulated data and actual spectral data show that the double-bird flock optimization algorithm iteration converges and can overcome the local minimum problem.Compared with similar algorithms, the three indices of abundance reconstruction error, average spectral angular distance and pixel reconstruction error of the algorithm are smaller, which shows that the proposed algorithm has high resolution of unmixing, good reconstruction effect of pixel, and can effectively improve the nonlinear unmixing accuracy of hyperspectral images.