异常检测是高光谱遥感技术应用的一个重要方向.然而随着高光谱数据量的增大, 实时处理成为高光谱异常检测方法所面临的主要问题.基于此, 文中提出了一种新型的高光谱图像实时异常检测方法.随着数据的实时下行传输, 该异常算子仅仅利用了待检测像元之前已获取的所有像元信息, 而并没有用到尚未获取的像元信息, 使得数据边传输边处理成为可能;同时, 利用卡尔曼滤波器的递归思想, 用Woodbury引理从上一时刻的状态更新目前信息, 避免了重新计算历史信息及存储所有像元, 在大大缩短算法运行时间的同时, 大大降低了所需的存储空间.接收机特性曲线显示, 与传统异常检测算法相比, 这种新型实时算法可获得几乎相同的检测精度.在不影响检测效果的前提下, 时间复杂度曲线和算子运行时间可显示提出算法的时效性.与此同时, 提出的的状态更新公式不需要重新计算已有像元信息, 因此只需两个存储单元存储前一时刻的状态(协方差矩阵或相关矩阵)以及当前的新像元信息, 从而大大降低了算法所需的存储空间.

Anomaly detection is one of the most important applications in hyperspectral imagery. Real-time processing is the main issue we are facing due to the large data set. Real time causal processing algorithms were developed to perform anomaly detection. It is an innovational kalman filtering based processing by using Woodburys identity to update information which provides the pixel currently being processed without re-processing previous pixels. Experimental results demonstrated the proposed algorithm significantly improves processing efficiency in comparison with conventional anomaly detection without real time causal processing.