联合空间预处理与谱聚类的协同稀疏高光谱异常检测

Joint Spatial Preprocessing and Spectral Clustering Based Collaborative Sparsity Anomaly Detection for Hyperspectral Images

针对利用稀疏表示进行高光谱图像异常目标检测效率不高的问题,基于高光谱图像成像原理和图像结构,充分利用高光谱图像的空间特性和光谱特性,并在它们之间建立协同处理机制,提出了联合空间预处理与谱聚类的协同稀疏高光谱图像异常目标检测算法。该算法首先对高光谱图像空间特性进行分析,并结合光谱特性进行空间预处理,使得处理后的高光谱图像更易于异常目标的检测;利用建立在谱图划分思想基础上的谱聚类方法进行波段子集划分,谱聚类方法具有收敛于全局最优解、聚类速度快的特点;利用提出的新的空间和光谱协同稀疏差异指数方法对每个子集进行异常目标检测,该协同稀疏方式充分考虑了高光谱图像的空间特性和光谱特性,通过对每个波段子集检测结果进行叠加,得到最终异常检测结果。利用真实的AVIRIS高光谱图像和合成的高光谱图像对算法进行仿真实验和结果分析,结果表明该算法具有稳健性,同时检测精度高,虚警率低。

In order to overcome the low efficiency of anomaly detection for hyperspectral images based on sparse representation, a joint spatial preprocessing and spectral clustering based collaborative sparsity anomaly detection algorithm is proposed, which makes full use of the spatial and spectral properties of hyperspectral images, and establishes a cooperative processing mechanism between the spatial and spectral properties, according to the imaging principle and structure of the hyperspectral imagery. The spatial properties of the hyperspectral images are analyzed, and the spatial preprocessing is combined with the spectral properties, which makes the anomalous targets in hyperspectral images easier to be detected. Then, the spectral clustering method based on spectrogram division is used to divide the band subsets, and the spectral clustering method has the features of convergence to the global optimal solution and fast speed. The anomalous targets in each band subset are detected with the proposed new space and spectral collaborative sparsity divergence index method. This collaborative sparsity method considers the spatial and spectral properties of the hyperspectral imagery. Final anomaly detection result is obtained by the superposition of the results of each band subset. The real AVIRIS and synthetic hyperspectral imagery data sets are used for simulations. Simulation results demonstrate that the proposed algorithm is robust, and has higher precision and lower false alarm probability.