基于多线性稀疏主成分的高光谱影像特征提取

Feature extraction of hyperspectral image based on multilinear sparse principal components

高光谱影像特征提取有助于提高高光谱数据的应用效率和精度。针对基于向量的特征提取算法无法充分利用高光谱影像立方体空间结构信息这一缺点,本文提出在所有张量模式中执行稀疏降维的多线性稀疏主成分分析(MSPCA)算法,以中国嘉兴典型村庄和美国内华达州Curprite矿区高光谱影像为原始数据,运用主成分分析(PCA)、空间主成分分析(SPCA)和多线性判别分析(MPCA)3种特征提取方法对比分析所提算法特征提取后的分类精度。结果表明,利用MSPCA进行特征提取得到的分类精度均优于其他方法,在两个试验区的总体分类精度分别达到96.36%和95.00%。

The feature extraction of hyperspectral image helps to improve the application efficiency and accuracy of hyperspectral data.Considering the disadvantage that vector based feature extraction algorithm could not make full use of the cube spatial structure information of hyperspectral image,the multilinear sparse principal component analysis(MSPCA)algorithm was proposed to perform sparse dimensionality reduction in all tensor modes.Based on the hyperspectral images of typical villages in Jiaxing,China,and the currite mining area in Nevada,USA,three feature extraction methods i.e.,principal component analysis(PCA),spatial principal component analysis(SPCA)and multi-linear discriminant analysis(MPCA),were used to compare and analyze the classification accuracy of the proposed algorithm after feature extraction.The results showed that the classification accuracy of feature extraction obtained by MSPCA was better than that of the other three methods,and the overall classification accuracy of the proposed algorithm in the two experimental areas was 96.36%and 95.00%respectively.