PCA与移动窗DWT特征提取-中国图象图形学报.DOC

中图法分类号：TP751 文献标识码：A 文章编号： PCA与移动窗小波变换的高光谱决策融合分类 叶 珍， 何明一 (西北工业大学电子信息学院 西安 710072) 摘 要：目的 高光谱数据具有较高的谱间分辨率和相关性，给分类处理带来了一定的困难。为了提高分类，本文提出了一种结合PCA与移动窗小波变换的高光谱决策融合分类算法。方法 首先，利用相关系数矩阵对原始高光谱数据进行波段分组；然后，利用主成分分析对每组数据进行谱间降维；再根据提出的移动窗小波变换法进行空间特征提取；最后，采用线性意见池（LOP）决策融合对多分类器的分类结果进行融合。结果 采用两组来自不同传感器的数据, 所提算法的分类精度和Kappa系数均高于已有的五种分类算法。与SVM-RBF算法相比，本文算法的分类精度高出了8%左右。结论 实验结果表明，本文算法充分挖掘高光谱图像的谱间-空间信息，在小样本情况下和噪声环境中具有良好的分类性能。 关键词：高光谱分类；主成分分析；小波变换；决策融合 PCA and Window Wavelet Transform for Hyperspectral Decision Fusion Classification Ye Zhen， He Ming-yi (Electronics and Information school, Northwestern Polytechnical University, Xi’an 710072, China) Abstract: Objective For hyperspectral data, high spectral resolution and correlation hinders the application of classification. To improve the classification accuracy, a hyperspectral decision fusion classification method based on principal component analysis (PCA) and windowed wavelet transform is proposed in this paper. Method Firstly， correlate coefficient matrix is used to group original hyperspectral data. Then, PCA is employed to reduce spectral dimensions of each group data. Following this, the proposed windowed wavelet transform method is used for spatial feature extraction. Finally, linear opinion pool (LOP) is employed to fuse the classification results from multi-classifiers. Result The proposed algorithm can obtain higher classification accuracy and Kappa coefficient than the existed five algorithms by using two hyperspectral datasets from different sensors. The classification accuracy of the proposed algorithm outperforms the classification accuracy of SVM-RBF about 8%. Conclusion Experimental results show that the proposed method can explore spectral-spatial information from hyperspectral imagery, improve the classification accuracy efficiently and provide outstanding classification performance under small-sample-size conditions as well as noise environments. Key words: Hyperspectral classification; principal component ana