提出了一种基于LKRX检测器的实时异常检测算法.利用局部因果滑动阵列窗, 使检测系统保持因果性.根据卡尔曼滤波器的递归思想, 利用Hermitian矩阵分块求逆引理和Woodbury引理, 将LKRX算法中核协方差矩阵以及其逆矩阵以递归方式更新, 避免了数据的重复计算和逆矩阵的求解, 大大降低了算法复杂度.通过真实数据进行实验, 结果表明, 与LKRX算法相比, 实时LKRX算法在保持相同检测精度的同时, 消耗更少的计算时间; 而与实时RX算法相比, 实时LKRX算法能够检测到更多的异常目标.

LKRX detector-based hyperspectral real-time anomaly detection algorithm was proposed. Using local causal sliding array window, the causality of detection system is remained. According to Kalman filter, by using Hermitian lemma and Woodbury’s identity, the kernel covariance matrix and its inverse in KRX algorithm are updated recursively. This thereby leads to low computational complexity. Experimental results demonstrated that real-time KRX detector consumes less time in comparison with KRX detector by keeping the same detection performance, which detects more anomalies.