摘要
支持向量机(Support Vector Machines,SVMs)的分类结果严重受限于模型参数的选择.很多学者都在研究调参问题,但目前还没有一个行之有效的方法,常用的办法是网格搜索,一个近似选择参数的方法.以噪声(光暗,有遮挡)图像分类为背景,以正则化支持向量机(Regularized SVM,RSVM)为分类器,研究了偏微分方程组(Partial Differential Equations,PDEs)对RSVM模型中参数选择的影响.实验结果表明通过PDEs的进化可以弱化参数的影响,甚至不需要考虑调参.
The classification results of support vector machines (SVMs)were heavily limited by model parameters selection.Parameter selection problem was studied by many researchers,but at present,there is no effective method.The grid searching method was usually used,which is only an approximate method.This paper is devoted to research the influence of partial differential equations (PDEs)for parameters selection with regularized SVM (RSVM)and a noisy image (illumination,shade)classification task.Experiment results indicate that the evolution of PDEs can weaken the parameters influence,even no need to consider this problem.
作者
江珊珊
范丽亚
JIANG Shan-shan;FAN Li-ya(School of Mathematical Sciences,Liaocheng University,Liaocheng 252059,China)
出处
《聊城大学学报(自然科学版)》
2019年第3期36-45,共10页
Journal of Liaocheng University:Natural Science Edition
基金
国家自然科学基金项目(11801248)
山东省自然科学基金项目(ZR2016AM24)资助
关键词
支持向量机
偏微分方程
参数选择
进化次数
图像分类
support vector machines
partial differential equations
parameters selection
evolutional times
image classification