基于特征贡献选择的遥感影像变化检测方法研究

基金项目: 中国地震局地震研究所所长基金资助项目(IS201116002,IS201056076)

详细信息

Remote Sensing Image Change Detection Method Based on Selection of Feature Contribution

¹Institute of Seismology,China Earthquake Administration;²School of Remote Sensing and Information Engineering,Wuhan University

Funds: 中国地震局地震研究所所长基金资助项目(IS201116002,IS201056076)

摘要: 根据决策树算法中的信息增益率概念,提出一种基于特征贡献度的特征选择方法。该方法通过累加统计各项特征在类别分析中的贡献,确定特征贡献度并作为权值代入特征空间距离计算公式,实现高维特征的降维和去相关。实验表明,同源影像间的特征选择可以有效地提高变化检测精度。
- 信息增益率 /
- 特征贡献度 /
- 特征选择 /
- 变化检测
Abstract: Based on the concept of information gain ratio in the decision tree algorithm, this paper has proposed a novel feature selection method using the attribute contribution. The contributions of attribute are obtained by accumulating in the category analysis. And then the feature space distance formula can be calculated using the feature contribution as weights. The feature contribution helps to reduce the dimension of the high-dimensional feature. The proposed method is tested in the remote sensing image change detection. The experimental results show that the proposed feature selection method can effectively improve the accuracy of the homologous image change detection results.
- information gain ratio /
- feature contribution /
- feature selection /
- change detection

[1]	孙淮宁,胡学钢. 一种基于属性贡献度的决策树学习算法[J]. 合肥工业大学学报(自然科学版). 2009(08)[2] 王圆圆,李京. 基于决策树的高光谱数据特征选择及其对分类结果的影响分析[J]. 遥感学报. 2007(01)[3] 孙微微,刘才兴,田绪红. 基于增益的数据样本分布描述方法[J]. 计算机应用. 2005(05)[4] 舒宁. 关于多光谱和高光谱影像的纹理问题[J]. 武汉大学学报(信息科学版). 2004(04)[5] 刘永军. 大数据集的属性选择算法的研究与实现[D]. 东北大学 2005