一种新的特征提取方法

Feature Extraction Based on Relief-F Algorithm and PCA Transform

针对影像分类识别中,属性特征过多不但会造成维数灾难,而且会影响分类精度的问题,该文采用基于Relief-F算法的主成分分析(PCA)变换特征提取方法解决特征降维问题。首先采用Relief-F算法进行特征选择,剔除无效特征;然后进行PCA变换减少特征之间的相关性,降低特征维数。定量分析与实验结果表明:Relief-F算法进行特征选择,能有效提高分类精度;进行PCA变换后,进一步降低了特征的维度;Relief-F算法与PCA变换相结合能实现较好的实验效果。

In image classification and recognition, too many attributes can cause dimension disaster and affect the classification accuracy. In order to solve the problem of feature dimension in image classification, this paper uses the method of feature extraction based on Relief-F algorithm and PCA transform. Firstly, eliminate invalid features using the method of feature selection based on Relief-F algorithm. Then, reduce the correlation between the features by PCA transform in order to reduce feature dimensions. Experimental results show that the Relief-F algorithm can effectively improve the classification accuracy and the PCA transform can further reduce the feature dimensions.