论文部分内容阅读
利用扩散映射所诱导出的扩散几何坐标对高光谱影像低维可视化表示,这种非线性维度约简的表示方法能够得到高光谱影像紧凑而富含信息量的可视化表示结果。通过对海量高光谱影像中每个高光谱观测向量进行局部搜索,仅考虑局部的邻接性和局部的相似度,构造出该高光谱影像对应的近邻图;对近邻图进行合适的规范化,得到该高光谱影像对应的扩散算子,相当于利用该扩散算子对高光谱特征空间模拟出马尔可夫随机游走.因此这样的构造较好地把握了高光谱影像内蕴的几何信息,与传统的基于主成分分析的线性降维表示方法相比,由扩散算子的特征分解所诱导出的扩散几何坐标能够给出更好的表示效果,富含更多的信息.对于大尺度的全景高光谱影像,利用构造“骨干”扩散几何坐标系的方法,其计算的时间复杂性和空间需求都是可接受的.实验也表明,选择合适的对称化方法规范扩散算子对于最终的高光谱影像表示有重要的影响.
Using the diffusion geometry coordinates induced by diffusion mapping, a low-dimensional visualization of hyperspectral images is presented. This non-linear dimensionality reduction method can obtain a compact and informative visualization result of hyperspectral images. Mass hyperspectral image by hyperspectral observation vector for each local search, considering only partial abutment and local similarity, the hyperspectral image constructed corresponding neighborhood graph; to be suitably normalized neighborhood graph to give the hyperspectral image corresponding to the diffusion operator, corresponding to the diffusion operator using hyperspectral simulated feature space Markov random walk. such a configuration thus a better grasp of the intrinsic geometry of hyperspectral image, the conventional Compared with the linear dimensionality reduction method based on principal component analysis, the diffusion geometric coordinates induced by the characteristic decomposition of diffusion operator can give better representation and contain more information.For the large-scale panoramic height spectral image, using the configuration “backbone ” geometric coordinates diffusion method, which calculates the spatial complexity and time requirements are acceptable. experiments also show, select an appropriate method specification symmetry diffusion operator for the final high Spectral imagery has a significant impact.