采用多项式递归核的高光谱遥感异常实时检测算法

Real-Time Anomaly Detection Algorithm for Hyperspectral Remote Sensing by Using Recursive Polynomial Kernel Function

高光谱遥感目标检测是遥感信号处理领域的热点问题,基于核机器学习的KRX算法能充分利用高光谱波段间的非线性光谱特性,在原始光谱的特征空间进行探测,能够获得较好的检测效果。针对KRX算法检测过程计算复杂、不能满足快速处理要求的缺陷,引入了卡尔曼滤波器的递归思想,提出了一种核递归的高光谱异常目标检测算法。从光谱分析的角度,应用Woodbury引理从上一时刻的状态迭代更新当前像元的Gram核矩阵,避免了高维矩阵数据重复计算。实验结果表明,与传统RX、因果RX和KRX等算法相比,在检测精度有所提高的同时,大大缩短了算法检测时间,提高了异常目标检测效率。

Hyperspectral target detection is a great deal of attention in the field of remote sensing signal processing. The KRX algorithm based on kernel machine learning can make full use of nonlinear spectral characteristics among hyperspectral bands. Therefore, it can get better detection results in the original spectral feature space. Aimed at the defect that the complexities of KRX algorithm is high in calculating the detection process and unable meet the requirement of rapid processing. A real-time anomaly detection method is proposed based on recursive kernel function. The recursive thought of Kalman filter is introduced, which puts forward a nuclear recursive hyperspectral anomaly target detection algorithm. From the perspective of spectral analysis, with Woodbury＇s lemma, the kernel matrices can be updated by the kernel matrices of last pixel. It avoids repeat computation of high-dimensional data matrices. Experimental results show that the accuracy of anomaly detection is improved and testing time of the algorithm is reduced at the same time when compared with the traditional RX, causal RX and KRX algorithm.