摘要
针对高光谱异常检测中临近异常像素相互干扰和背景地物复杂的问题,提出基于局部投影可分离的高光谱图像异常检测算法。在归一化的数据中,将待测像素光谱作为参考光谱,构造目标子空间,然后把邻域背景像素投影到该子空间,用投影后向量模值构造异常度计算式。最后将检测到的异常与全局主要背景地物进行比对,消除部分虚警。利用HyMap高光谱数据进行仿真实验结果表明,本文算法具有克服背景复杂性和干扰点的影响,尤其对异类干扰点的抑制效果更佳。
Aiming at the interference of close outliers and the complexity of background features, a new anomaly detection method based on separable local projection in hyperspectral imagery is proposed. After normalizing the data, the test pixel spectrum is selected as the reference spectrum to build the target subspace. Then, the background pixels in close areas are projected onto this subspace and the formula of the abnormal degree is acquired through the modulus of the projection vec- tors. Finally, comparing the anomalies with the main surface features, parts of the false alarms are eliminated. The experi- ments were conducted on HyMap hyperspectral data and the results show that the proposed algorithm overcomes the impact of background complexity and interference pixeis, especially in situations where the interference pixels and test pixel are in different classes.
出处
《中国图象图形学报》
CSCD
北大核心
2013年第5期558-564,共7页
Journal of Image and Graphics
关键词
高光谱图像
异常检测
局部投影
归一化处理
hyperspectral images
anomaly detection
local projection
normalized process