基于四阶累积量的波段子集高光谱图像异常检测

Anomaly detection of hyperspectral image for band subsets based on fourth order cumulant

针对由于高光谱图像光谱和空间分布的复杂性导致核RX算法检测性能不高这一问题,提出了基于四阶累积量的波段子集非线性异常检测算法。首先先依据各相邻波段间的相关系数,将原始图像数据划分为多组波段子集;然后,利用主成分分析(PCA)构造的正交子空间对各波段子集进行背景抑制,得到图像误差数据;在此基础上,再次利用PCA提取各波段子集的特征信息,使异常目标信息集中于前面几个波段;最后,提取各子集主成分中含有最大四阶累积量值的波段,构成最优波段子集,并与核RX算法结合进行异常检测。利用真实的AVIRIS高光谱图像对算法进行仿真,结果表明,本文算法检测精度高,虚警率低,性能明显优于核RX算法。

The anomaly detection performance of kernel RX is low as a result of the complexity of spectral and spatial distributions of hyperspeetral images. Aiming at the problem, this paper proposes an algorithm of band subsets anomaly detection for hyperspectral image based on fourth order cumulant. Firstly, the algorithm divides the original hyperspectral image into subsets in low dimensions according to the correlation coefficient between spectral bands. Then, the error image data are achieved after background interferences are suppressed for subset bands using the orthogonal sub-spaces constructed by principal component analysis. Based on the data, the feature information of all band subsets is extracted by using the principal component analysis, which makes the information of anomaly target concentrated on the previous bands. At last, the optimal band subsets are achieved by fourth order cumulant of principal com- ponent in band subsets,and the anomaly detection is carried out combined with the kernel RX. The re sults show that the proposed algorithm has higher precision and lower false alarm probability, and it greatly outperforms the classical kernel RX algorithm.