摘要
支持向量机的高光谱图像分类中,单核函数存在局限性。为了提高分类器的分类精度和支持向量机模型的泛化能力,利用高斯径向基核和多层感知核进行凸组合构造复合核函数支持向量机,证明了该函数满足作为核函数的判决Mercer条件,并进一步将凸组合核函数支持向量机应用到高光谱图像分类中,完成了建模和实验验证。实验结果表明,凸组合核函数具有较好的鲁棒性,且该类支持向量机的分类精度和KAPPA系数较单核SVM均得到了有效的提高,是一种解决多分类问题行之有效的分类器。
In support vector machine（ SVM） hyperspectral image classification,monocyte function has its limitation.In order to improve the classifier accuracy and generalization ability of SVM model,a complex kernel function SVM using the convex combination of radial basis function kernel and sigmoid kernel was constructed,and it proves that the function satisfies a judgment called Mercer condition as a kernel function.Then,the convex combination of kernels SVM was applied to hyperspectral image classification,and the modeling and experimental validation were completed.The experimental results show that the convex combination kernel has better robustness.As the classification accuracy and KAPPA coefficient have been effectively improved compared to that of the single-core SVM,the new SVM is an effective solution to the problem of multi-classification.
出处
《激光与红外》
CAS
CSCD
北大核心
2016年第5期627-633,共7页
Laser & Infrared
基金
国家自然科学基金资助项目(No.61179036
No.61201379)资助
关键词
高光谱图像
支持向量机
核函数
凸组合
hyperspectral image
SVM
kernel fuction
convex combination
