随着光谱成像技术的发展, 高光谱异常检测在遥感图像处理中的应用越来越广泛。 传统RX异常检测算法忽略影像空间相关性, 而且由于没有经过有效数据降维, 运算耗费大, 对于高光谱数据有效性不高。 高光谱影像在空间和光谱上符合高斯-马尔科夫模型。 通过建立马尔科夫参数能够直接计算协方差矩阵的逆矩阵, 避免了高光谱海量数据的庞大计算。 提出一种基于三维高斯-马尔科夫随机场模型的改进RX异常检测算法。 该方法用高斯-马尔科夫随机场模型模拟高光谱影像数据, 用最大似然近似法估计高斯-马尔科夫随机场参数, 由高斯-马尔科夫随机场参数直接构造检测算子, 并以待检测像元为中心设置局部优化窗口, 称为马尔科夫检测窗。 取窗口内数据计算均值向量和协方差逆矩阵, 得到中心像元的异常度, 通过移动窗口进行逐像元检测。 应用AVIRIS高光谱数据对传统RX算法、 高斯-马尔科夫模型背景假设异常检测算法和该算法进行了仿真实验对比。 结果表明, 该算法能够有效提高高光谱异常检测效率, 降低虚警率。 运行时间较传统RX算法提高了45.2%, 体现出更好的计算效率。

With the development of spectral imaging technology, hyperspectral anomaly detection is getting more and more widely used in remote sensing imagery processing. The traditional RX anomaly detection algorithm neglects spatial correlation of images. Besides, it doesnot validly reduce the data dimension, which costs too much processing time and shows low validity on hyperspectral data. The hyperspectral images follow Gauss-Markov Random Field (GMRF) in space and spectral dimensions. The inverse matrix of covariance matrix is able to be directly calculated by building the Gauss-Markov parameters, which avoids the huge calculation of hyperspectral data. This paper proposes an improved RX anomaly detection algorithm based on three-dimensional GMRF. The hyperspectral imagery data is simulated with GMRF model, and the GMRF parameters are estimated with the Approximated Maximum Likelihood method. The detection operator is constructed with GMRF estimation parameters. The detecting pixel is considered as the centre in a local optimization window, which calls GMRF detecting window. The abnormal degree is calculated with mean vector and covariance inverse matrix, and the mean vector and covariance inverse matrix are calculated within the window. The image is detected pixel by pixel with the moving of GMRF window. The traditional RX detection algorithm, the regional hypothesis detection algorithm based on GMRF and the algorithm proposed in this paper are simulated with AVIRIS hyperspectral data. Simulation results show that the proposed anomaly detection method is able to improve the detection efficiency and reduce false alarm rate. We get the operation time statistics of the three algorithms in the same computer environment. The results show that the proposed algorithm improves the operation time by 45.2%, which shows good computing efficiency.