一种用于高光谱遥感影像分类的改进多类支持向量机

将支持向量机(SVM)用于高光谱遥感影像分类的研究,采用决策边界特征提取(DBFE)算法对高光谱影像进行维数约简,以径向基函数(RBF)作为SVM模型的核函数,把混沌优化搜索技术引入到PSO算法中,以基本PSO算法为主体流程,对种群中最好的粒子进行给定步数的混沌优化搜索,以改进基本PSO算法进化后期收敛速度慢、易陷入局部极小值的缺陷。利用改进的混合粒子群优化算法(PSO)来实现SVM模型参数的自动选择,继而构建了一种参数最优的粒子群优化支持向量机(PSO-SVM)多类分类模型。选用220波段的AVIRIS高光谱遥感影像进行了分类试验。结果表明,与采用基于留一法(LOO)网格搜索策略的传统SVM相比,改进后的PSO-SVM算法可以提高分类精度约8.8%。该方法对于小样本、非均衡条件下的遥感影像数据分类非常有效。

An Improved Multi-class Support Vector Machine for Hyperspectral Remote Sensing Imagery Classification

Support Vector Machine (SVM) is applied in the research on hyperspectral remote sensing image classification. Decision Boundary Feature Extraction (DBFE) algorithm is adopted as dimension reduction for hyperspectral image data. Radial Basis Function (RBF) is used as SVM kernel function. The chaotic optimization search technique was introduced into the Particle Swarm Optimization (PSO) algorithm. Its main idea is, within the basic PSO framework, to overcome disadvantages of low convergent speed and easily falling into local minimum by chaotic optimization search with predetermined number of steps for the best particle among a population. The improved PSO algorithm is proposed to automatically determine SVM model parameters, and the PSO based SVM (PSO-SVM) model was constructed. Experimental results of 220 bands AVIRIS hyperspectral data indicate that PSO-SVM algorithm, we proposed in this paper, can improve the overall classification accuracy about 8.8%, as compared to the standard SVM with Leave One Out(LOO) based grid search strategy. The proposed algorithm is very efficient for small samples and unbalanced remote sensing data cmassification.