综合纹理特征的高光谱遥感图像分类方法

摘要

A SVM classification methodology combined with Gabor texture feature for hyperspectral remote sensing image data is proposed. By introducing texture feature produced by Gabor filter group into SVM classification, the algorithm takes advantage of SVM that is proper for high dimensional data classification, and integrates spatial and spectral information during classification process. The experiments with OMIS (operative modular imaging spectrometry) real hyperspectral image data show that the algorithm improves classification performance with higher classification correctness, the classification results show more spatial consistency, and the method can effectively eliminate the influence of noise.%提出了一种基于Gabor滤波的高光谱遥感图像支持向量机(SVM)分类方法,通过将Gabor滤波器组产生的纹理特征引入SVM分类,不仅充分利用了SVM适于解决高维数据分类问题的优势,而且在分类过程中实现了空间结构信息和光谱信息的综合使用,有效利用了高光谱图像“图谱合一”的特性.采用中科院上海技术物理研究所研制的模块化成像光谱仪OMIS (operative modular imaging spectrometry)真实数据进行的实验,实验结果表明,该方法提高了分类效果,分类结果更具有空间连贯性,并且能有效地克服噪声的影响.