计算机科学与探索 ›› 2020, Vol. 14 ›› Issue (12): 2140-2149.DOI: 10.3778/j.issn.1673-9418.2002003

• 图形图像 • 上一篇    下一篇

基于低秩全变差正则化的高光谱异常检测方法

徐超,詹天明   

  1. 南京审计大学 信息工程学院,南京 211815
  • 出版日期:2020-12-01 发布日期:2020-12-11

Hyperspectral Anomaly Detection Method Based on Low Rank Total Variation Regu-larization

XU Chao, ZHAN Tianming   

  1. School of Information Engineering, Nanjing Audit University, Nanjing 211815, China
  • Online:2020-12-01 Published:2020-12-11

摘要:

高光谱遥感技术为探索地物提供了丰富的信息,为异常检测提供了更优质的数据源。但是在先验信息未知的情况下,进行异常检测仍是一项非常具有挑战性的工作。针对该问题,提出一种基于低秩和全变差正则化约束的高光谱数据异常检测方法。首先,对高光谱图像进行线性和非线性解混,得到两组丰度图像,将丰度图像与原高光谱图像进行融合。其次,根据背景区域在融合数据中的特征构建图像背景的字典,并建立图像的低秩表示模型。然后,由背景和异常目标各自特点,建立异常检测正则化模型。最后,对模型进行优化求解,得到异常检测结果。在真实高光谱数据中进行实验,实验结果表明该方法可获得较优的高光谱异常检测性能。

关键词: 高光谱数据, 异常检测, 低秩表示, 正则化

Abstract:

Hyperspectral remote sensing technology provides abundant spectral information for exploring objects and supplies a better data source for anomaly detection. However, anomaly detection is still a challenging task without any valuable prior information. Aiming at this problem, a hyperspectral anomaly detection method based on low-rank and TV regularization constraint is proposed in this paper. Firstly, the hyperspectral image is linearly and nonlinearly unmixed to generate two abundance maps, and these two maps are fused with the original hyperspectral image. Secondly, the spectral dictionary of background targets in hyperspectral image is constructed according to their features in the fused data, and a low-rank representation model of the image is generated. Thirdly, an anomaly detection regularization model is established according to the characteristics of normal and abnormal targets. Finally, the model is optimized to generate the anomaly detection result. Experiments are carried out in the real hyperspetral datasets, and the detection results demonstrate that the proposed method is able to achieve a promising anomaly detection performance.

Key words: hyperspectral data, anomaly detection, low-rank representation, regularization