基于高维几何特性的高光谱异常检测算法研究

提出了一种新的高光谱图像异常检测算法。作为一种多元数据集合,高光谱数据一般呈现出共超平面的几何特性,我们利用这一特点来求取垂直于超平面的法线矢量,并将数据投影到这一法线矢量方向,从而分离出异常点,达到异常检测的目的。本算法适合于对小目标的检测,且不需要先验的光谱信息。对算法的可行性进行了仿真并将它应用于高光谱数据,取得了较好的结果。

The Research of Anomaly Detection Based on High-Dimensional Geometrical Feature in Hyperspectral Imagery

A new method of hyperspectral anomaly detection is presented in this paper. As a kind of multivariate data, the points of hyperspectral data always stay in a hyper-plane in high-dimensional space. More researches show that the points of anomaly spectral and noise stay out of the configuration. This geometrical feature is useful for small targets detection. We can calculate the normal vector of hyper-plane and project all anomaly pixel vectors to the normal direction, thus we can segment the anomaly target from the hyper-plane. The method adapts to detect small targets in complicated scene. We design a simulation experiment to verify the algorithm and apply it to real data, and get satisfying results.