基于简约集支持向量机的高光谱影像分类

Reduced Set Based Support Vector Machine for Hyperspectral Imagery Classification

针对高光谱影像支持向量机分类的预测过程中需要花费大量计算时间的问题,提出了一种利用简约集算法提高高光谱影像分类预测效率的方法。采用径向基核函数,使用一对一构造多类支持向量机分类器,通过交叉验证网格搜索法对参数进行模型参数选择,并利用简约集算法来减少分类预测过程计算量。通过高光谱影像分类试验表明,保持支持向量机的泛化能力并不需要使用所有计算得到的支持向量,简约集算法能够在保持分类预测精度的同时,大大提高高光谱影像分类预测的速度。

Aiming at the problem of more computational time needed in hyperspectral imagery classification procedure based on support vector machine,a reduced set method was brought forward to heighten hyperspectral imagery classification efficiency.The radial basis kernel function was adopted,one-against-one decomposition algorithm was used to construct multi-class Support Vector Machine classifier and cross validation grid search method was applied to select model parameter.The reduced set algorithm was also used to reduce the computational complexity of predication.Through hyperspectral imagery classification experiment it can be concluded that it does not need to use all support vectors to keep generalization ability of Support Vector Machine.The reduced set algorithm can improve hyperspectral imagery classification predicative efficiency highly and keep classification accuracy at the same time.