【摘要】 针对传统基于像素的多光谱遥感图像分类方法存在的"麻点"现象、采样成本高等问题,提出了一种基于图论分割的非监督分类方法,首先采用基于图论的分割算法,按局部邻近相似像素点分割成若干子区域,再以分割后子区域为基本单元,整体进行模糊C均值聚类,最终实现对多光谱图像的非监督分类.实验证明,该方法结合了局部邻近像素点的相互关系以及相似区域的整体特征,有效解决了麻点问题,具有较高的分类精度和算法效率,降低了采样成本.更多还原

【Abstract】 To solving the noisy points and high cost problems of pixel-based multispectral image classification,a hybrid unsupervised approach with graph-based segment and fuzzy c-means clustering was presented.First,based on the relationships among neighboring pixels,image was segmented into groups of sub-regions using the graph-based algorithm.Then according to the global feature vector of sub-region,the fuzzy c-means classifier was used to obtain the classification map.Experiments turn out that the proposed approach,which considers both relationships of neighboring pixels and global feature of sub-region,can achieve better accuracy and efficiency by comparing the result with pixel-based fuzzy c-means classification.更多还原