Abstract: According to the fact that high-dimensionality and redundant information between bands of hyperspectral data, the paper proposes a new classification method based on homogeneity dimensionality reduction and Combined Matching Pursuit（CMP） algorithm. Firstly, the proposed method segments hyperspectral image by mean shift algorithm and obtains homogeneous blocks, and builds dimensionality reduction mapping function of the image block by manifold learning. And then the method trains sparse least squares SVM for low-dimensional data, but the method proposes a  new sparse reconstruction method based on CMP algorithm for solving multiple iterations of Orthogonal Matching Pursuit（OMP） algorithm. By classification results of hyperspectral data can be drawn that the proposed method effectively improves the hyperspectral image classification accuracy.

Key words: hyperspectral image classification, homogeneity dimensionality reduction, sparse least squares SVM, Combined Matching Pursuit（CMP） algorithm

摘要： 针对高光谱数据维数高，波段间冗余信息大的问题，提出一种基于同质性降维和组合匹配追踪算法的高光谱图像分类方法。该方法首先利用均值漂移算法对高光谱图像进行分割得到同质性图像块，对同质性的图像块进行流行学习得到降维映射函数，然后由降维后的高光谱数据训练稀疏最小二乘支持向量机分类模型，为避免正交匹配追踪稀疏重构算法迭代次数多的缺点，提出一种基于组合匹配追踪的稀疏重构求解方法。通过高光谱数据的分类结果可以得出，该方法有效提高了高光谱图像的分类精度。

关键词: 高光谱图像分类, 同质性降维, 稀疏最小二乘支持向量机, 组合匹配追踪算法