基于背景抑制及顶点成分分析的高光谱异常小目标检测

Anomaly detection of hyperspectral imagery based on background restrain and VCA

高光谱图像异常小目标检测数据量大、信息提取困难.文中提出了一种不需要先验信息并且计算复杂度较低的快速检测算法——基于背景抑制及顶点成分分析(EVCA)的异常小目标检测.利用高光谱图像端元是单形体顶点这一特性,在抑制背景后的图像上提取目标端元,并结合光谱匹配技术完成目标检测.为了验证新方法的有效性,与不经过背景抑制的VCA算法及传统检测算法进行了比较.实验结果表明,该算法不需要先验信息,体现了较好的检测效果.

In order to solve the problem of mass data and hard extraction of information in the anomaly detection of hyperspectral imagery, this paper presents a fast detection algorithm that has lower complexity of computation and does not need prior knowledge--anomaly detection of hypcrspectral imagery based on background restrain and VCA （vertex component analysis）. By making use of the characteristic that the high spectral image＇ s endmembers are the vertexes of single figure, this algorithm extracts endmembers from the image after restraining background, and detects the anomaly target by combining the spectrum matching technology. But the anomaly target to detect usually has low probability and is badly affected by the noise from background. So we first extract the information of back- ground and use orthogonal projection to restrain it. The experiments show that this algorithm can improve the detec- tion performance.