The Classification Of Hyperspectral Remote Sensing Image Based On Hyper Sphere Multi-Class Support Vector Machine

Support vector machine is a new general machine learning method based on Statistical Learning Theory, can solve hughes phenomenon effectively. It based on statistical learning theory, concentrated several good technologies such as maximum margin hyper plane, Mercer kernel, convex quadratic programming, spare solutions and slack variables, which can overcome the shortages of traditional classifiers-local minimization, curse of dimension and overfitting. This classification there are still some shortcomings:high computation complexity, large memory demanding and very difficult to use for large scale remote sensing image data sets. This paper proposes that one should use Hyper Sphere Multi-Class Support Vector Machine (HSMC-SVM) and its improved Least Square Hyper Sphere Multi-Class SVM (LSHS-MCSVM). and semi-fuzzy kernel clustering LSHS-SVM (SFLSHS-MCSVM) to classify the hyperspectral remote sensing image, then good classification results are gotten.The major works and contributions of this paper are as follows:First, Machine learning, statistical leaning theory, the development of support vector machine are introduced. The theory and algorithm of support vector machine are summarized. And the theory of kernel function, parameters selection and other issues are discussed simultaneously. Analyzes various methods of Multi-class Support Vector Machines. Finally, through a description of the experimental data shows that SVM classification for hyperspectral remote sensing images can obtain good results.Second, this paper focuses on the principle and correlative conceptions about HSMC-SVM. For improving training speed and learning precision again, least square measure is introduced to HSMC-SVM and form the new learning machine-Least Square Hyper Sphere Multi-Class SVM (LSHS-MCSVM). Comparing to HSMC-SVM, LSHS-MCSVM exploits second norm in object function, replaces the inequation constrains with equation constrains and gets rid of the limitation of Lagrange multipliers. These differences cause faster multipliers scanning and optimization calculation in LSHSMC-SVM, so it has faster convergence speed than HSMC-SVM. Samples are preprocessed with semri-fuzzy kernel clustering to ensure that the ones near boundaries are selected and then used to train least square hypersphere support vector machine so as to improve its performance efficiently. Samples training uses sequential minimal optimization algorithm, under working set selection of Zoutendijk Feasible Direction Methods, the training speed is faster than empirical working set selection.Third, through PHI hyperspectral remote sensing images as experimental data obtained after the classification of overall accuracy and kappa coefficient, demonstrates that HSMC-SVM and its improved algorithms are feasible and valid. Through analysis that based on HSMC-SVM of hyperspectral remote sensing image classification algorithm has good effect, the relative standard classification algorithm of support vector machine, the classification of speed increased significantly, and good classification results are gotten.