摘要
支持向量机分类方法存在惩罚系数需要交叉验证获取、训练时间较长、支持向量个数随着训练样本数量的变化而变化,以及稳定性和稀疏性较差等问题。针对这些问题,提出了一种基于输入向量机的高光谱影像分类算法。该算法在核逻辑回归模型的基础上,采用前向贪心算法选择训练样本中的输入向量来进行模型的训练,达到稀疏的目的,提高影像的分类精度和分类效率。通过PHI和OMIS两组高光谱影像分类实验,结果表明基于输入向量机分类算法具有稳定性好、稀疏性强的优点。
Some deficiencies still exist in the classification algorithm based on support vector machine, for exam- ple, penalty coefficient is got by cross validation; training time is longer; the number of support vectors changes a- long with the training sample size, resulting in poor stability and sparsity. In order to overcome this problems, a classification algorithm for hyperspectral images based on import vector machine was proposed. On the basis of nu- clear logistic regression model, the greedy forward method is used to choose import vectors from training samples for the model training, so as to achieve sparsity and improve the efficiency and accuracy of classification. Through two groups of classification experiments with the PHI and OMIS hyperspectral images, the results show that the classification algorithm based on import vector machine has an advantage of hatter stability and stronger sparsity.
出处
《测绘科学技术学报》
CSCD
北大核心
2015年第4期379-383,共5页
Journal of Geomatics Science and Technology
基金
国家自然科学基金项目(41201477)
关键词
高光谱影像
输入向量机
支持向量机
分类
稀疏
hyperspectral image
import vector machine
support vector machine
classification
sparsity