Research On Hyperspectral Image Demixing And SVM Classification Parameter Optimization

In recent decades,with the rapid development of human beings in electronic technology,optical technology and space technology,remote sensing technology has also made great progress.Hyperspectral remote sensing technology is widely used in disaster assessment,military investigation,mineral analysis and other fields.Hyperspectral image unmixing and supervised classification are important components in hyperspectral image processing.Domestic and foreign scholars have made some progress in image demixing and image classification based on Support Vector Machine(SVM).However,the traditional de-mixing algorithm has high complexity and low efficiency,and the SVM-based classification accuracy is highly dependent on parameter selection,resulting in the final classification result is not very satisfactory.Therefore,a set of hyperspectral image preprocessing algorithms that can guarantee both real-time and classification accuracy are urgently needed.This paper studies two aspects from image demixing and SVM classification parameter optimization.Aiming at the low real-time problem of image de-mixing algorithm,a new fast linear de-mixing algorithm is proposed,The problem of slow convergence of improved particle swarm optimization algorithm for classification parameter optimization is proposed.Particle swarm algorithm for immune algorithm.The main research contents are as follows:Firstly,a new fast linear demixing algorithm(FLD)using Schmidt orthogonalization for end-selection and image de-mixing is designed to perform end-selection of images after wavelet-PCA transform denoising and dimension reduction.And unmixed.Both the endmember selection algorithm and the image demixing algorithm are based on the basic theory of Schmidt orthogonalization,so that the two algorithms can be run in the same system,which greatly simplifies the algorithm complexity and improves the real-time performance.The experimental results show that the algorithm has great advantages in real-time performance under the condition of ensuring accuracy.Secondly,the Immune Cross-Mutation Particle Swarm Optimization(ICMPSO)algorithm is proposed to optimize the parameters of the support vector machine(the penalty coefficient c and the radial basis kernel function parameter g).Compared with the common algorithm such as adaptive mutation particle swarm optimization algorithm,the algorithm has improved the effect effectively,and ensures stable convergence without falling into local optimum.The experimental results show that the algorithm can improve the classification accuracy by about 3%.Finally,the hyperspectral image obtained by FLD algorithm is brought into the SVM classifier for classification,and the ICMPSO algorithm is used to find the optimal parameters in SVM parameter selection.The experimental results show that the five categories of the 17 kinds of features of the hyperspectral image can be better identified,and the classification accuracy is greatly improved compared with the traditional algorithm.