| 特征贡献度与PCA结合的遥感影像分类特征选择优化方法研究 |
| |
| 引用本文： | 孙俊娇,王萍,张英,冯志贤,桑会勇.特征贡献度与PCA结合的遥感影像分类特征选择优化方法研究[J].测绘与空间地理信息,2018(1):49-54. |
| |
| 作者姓名： | 孙俊娇 王萍 张英 冯志贤 桑会勇 |
| |
| 作者单位： | 山东科技大学 测绘科学与工程学院,山东 青岛,266590
中国测绘科学研究院,北京,100830
兰州交通大学 测绘与地理信息学院,甘肃 兰州730070;城市空间信息工程北京市重点实验室,北京100830 |
| |
| 基金项目： | 专题性地理国情监测，地理国情监测国家测绘地理信息局重点实验室开放基金，2017年中国测绘科学研究院基本科研业务费项目"基于亚像元分析的洪水动态监测关键技术"，城市空间信息工程北京市重点实验室经费资助项目 |
| |
| 摘 要： | 面向对象遥感影像分类过程中,特征选择是保证分类精度和提高分类速度的关键因素。本文针对高分影像特征过多造成维度灾难、无法取舍有效特征导致低分类精度等问题,提出了一种基于特征贡献度与主成分分析(PCA)结合的特征选择优化方法,定量分析并提取影像特征。本文首先利用特征贡献度进行特征选择,提取有效特征;然后进行PCA变换消除特征间相互影响,降低维度,将提取的143个影像分类特征经选择与变换至20个主成分特征,最终优化的特征在神经网络(ANN)、K最近邻法(KNN)和支持向量机(SVM)三种分类实验结果中的总精度分别提高了10.56%、7.78%和6.11%,实现了较好的分类效果,说明优化的特征选择方法不仅大大降低了特征维度,减少了后端分类计算量,同时有效提高了分类精度。
|
| 关 键 词： | 贡献度 主成分分析 特征选择 遥感影像分类 contribution degree principal component analysis feature selection remote sensing image classification |
An Improved Method for Feature Selection of the Remote Sensing Image Classification Based on Feature Contribution and PCA Transform |
| |
| SUN Junjiao,WANG Ping,ZHANG Ying,FENG Zhixian,SANG Huiyong.An Improved Method for Feature Selection of the Remote Sensing Image Classification Based on Feature Contribution and PCA Transform[J].Geomatics & Spatial Information Technology,2018(1):49-54. |
| |
| Authors: | SUN Junjiao WANG Ping ZHANG Ying FENG Zhixian SANG Huiyong |
| |
| Abstract: | In the process of object-oriented remote sensing image classification, feature selection is the key factor to ensure the classi-fication accuracy and improve the classification rate. In this paper, an optimized method of feature selection based on the combination of feature contribution and principal component analysis ( PCA) is proposed to quantitatively analyze and extract image features, which can be used to solve the problem of dimension disaster caused by too many features and solve the problem of low classification accuracy caused by the inability to choose the effective feature. The feature selection model based on feature contribution is used to extract the effective features, and then the principal component analysis ( PCA) is used to eliminate the mutual influence between the features, fi-nally, the extracted 143 image classification features are selected and transformed to 20 principal component features. In the neural network ( ANN) and K nearest neighbor ( KNN) and support vector machine ( SVM) classification experiment, the classification re-sults of feature extraction contribution model combined with the PCA transform of the total accuracy was increased by 10. 56％, 7. 78％ and 6. 11％, what achieved better classification results. This results shows that the optimized feature selection method not only greatly reduces the feature dimension, reduces the calculation quantity of the back end, but also improves the classification accuracy. |
| |
| Keywords: | |
| 本文献已被 CNKI 万方数据 等数据库收录！ |
|
