河南城建学院学报Issue(3)：51-55,60,6.DOI:10.14140/j.cnki.hncjxb.2015.03.012
训练样本数量选择和总体分类精度的关系研究
Relationship between number of training samples selection and accuracy of overall classification
王春来 1张森原 1崔璐 1葛玉停 1张金禄 1张淼泓2
作者信息
- 1. 河南黄河勘测信息工程院，河南郑州450045
- 2. 黄河水文勘察测绘局，河南郑州450000
- 折叠
摘要
Abstract
The research objects between the classification based on the feature per -parcel and pixel are dif-ferent;there is a big difference in the choice of the number of training samples .In this paper , using high-resolution remote sensing image data and the classifier of SVM , the experiment showed that the accuracy can reach a high level based on feature per -parcel classification when the number of samples reach to 6-8 times of the band numbers and enter a stable phase;but the accuracy can reach a high level based on pixel when the number of samples reach to 24-30 times of the band numbers .关键词
监督分类/特征基元/训练样本Key words
supervised classification/feature per-parcel/training samples分类
信息技术与安全科学引用本文复制引用
王春来,张森原,崔璐,葛玉停,张金禄,张淼泓..训练样本数量选择和总体分类精度的关系研究[J].河南城建学院学报,2015,(3):51-55,60,6.
