Hyperspectral Image Unmixing Via Bionic Intelligent Optimizaton With Multi-Swarm Cooperaton

Hyperspectral remote sensing imagery are composed of images of dozens or even hundreds of bands in the same location.And they contain not only much spatial information,but detailed bands information.Because of the high spectral resolution,the level of spatial resolution of hyperspectral imagery are limited to ensure that contains low noise intensity.Due to the low spatial resolution,mixing pixels are commonly found in hyperspectral imagery.Different types of ground materials may be in the same pixel,and the spectral features are mixed badly,which has a great impact on the extraction of ground materials.In order to obtain a more accurate map distribution,hyperspectral unmixing is the first process to determine the various types of ground materials and each proportion of every pixel.As an emerging artificial intelligence method,the bionic intelligent optimization method has shown great advantages in solving complex optimization problems.Such an approach doesn't require prior assumptions or prior knowledge with gradients,Hessian matrices,or probability distributions.Therefore,there is a great potential for applying the bionic intelligent optimization method to hyperspectral unmixing.In this paper,we aim to improve the accuracy of hyperspectral unmixing,and use the emerging bionic intelligent optimization method to solve this problem suitably.The primary works of this thesis are as follows:(1)Aiming at the specific problem of unmixing,a new biomimetic intelligent optimization method,named,crossover double particle swarm optimization method is proposed to estimate the abundance of hyperspectral remote sensing imagery.Based on multilinear mixing model,the reconstruction error is used as objective function,and the nonlinear unmixing problem is further transformed into the optimization problem.Then the abundance matrix is estimated by the proposed method.Two types of new strategies were embed,including double particle swarms crossover and swarm re-initialization,respectively.Finally,experiments were performed on synthetic hyperspectral data and real hyperspectral data.The experimental results show that the proposed method can effectively improve the accuracy of hyperspectral unmixing.(2)A hyperspectral image unmixing method with dynamical self-adaption multi-swarm cooperation particle swarm optimization is proposed.Based on the crossover double particle swarm optimization method,a mechanism of judging the nonlinear mixing degree of an individual pixel is introduced.According to the degree of nonlinear mixing,the original mode of double swarm optimization is substitute for a new mode aboutmulti-swarm optimization pixel-by-pixel unmixing.For the pixel with a high degree of nonlinearity,a relatively large number of swarm are used for unmixing,and for the pixel with linearity or a lower degree of nonlinearity,a small amount of swarm are used for unmixing.Not only further improves the accuracy of unmixing,the proposed method also greatly improves the efficiency of mixing.The experimental results show that the proposed method has good and stable unmixing performance.