基于光谱角匹配加权的高光谱图像异常检测

Anomaly detection for hyperspectral image based on weighted spectral angle match

针对高光谱背景中存在异常和噪声的问题,提出了一种基于光谱角匹配(SAM)加权的核RX异常检测算法。首先对图像背景像元进行K-均值聚类,得到不同类背景对应的聚类中心,然后计算背景像元与聚类中心的光谱角余弦,选出较纯净的背景作为新背景,最后新背景中的每个像元将自己的光谱角信息作为权值,构造加权核RX异常检测算子,通过加权削弱了残留其中的异常和噪声的干扰。为验证算法的有效性,利用真实的AVIRIS和ROSIS-03遥感器采集高光谱数据进行了仿真实验,结果表明与对比算法相比,所提算法对潜在的异常具有较强的抑制能力,提高了检测精度。

In order to overcome the problem that hyperspectral image background samples contain anomalous pixels and noise, a kernel RX anomaly detection algorithm based on weighted spectral angle match （SAM） was proposed. Firstly, k-means clustering was performed on the background pixels of the image to obtain the cluster centers, then the spectral angle cosine of the background pixels and the cluster centers was calculated, the pure background was selected as the new background. Each pixel in the new background will own its spectral angle information as the weight, which is given to every background pixel to construct weighted kernel RX anomaly detector to weaken the interference of the residual outliers and noise. To validate the effectiveness of the proposed algorithm, experiments were conducted on real hyperspectral data from AVIRIS and ROSIS-03 remote sensor. The results show that by comparison with the compared algorithms, the proposed algorithm has strong suppression ability against potential outliers and can improve the detection accuracy.