基于端元提取的超光谱图像目标检测算法  

An Endmember Extraction Based Target Detection Algorithm for Hyperspectral Image

作　　者：张文希[1] 郑茂[2] 李纲[2] 

机构地区：[1]长沙学院电子与通信工程系,长沙410003 [2]国防科学技术大学电子科学与工程学院,长沙410073

出　　处：《电光与控制》2010年第8期45-48,共4页Electronics Optics & Control

摘　　要：针对超光谱图像中目标检测问题,提出了一种基于端元提取的超光谱图像目标检测算法。该算法在未知任何先验信息条件下,对超光谱数据进行基于噪声调节的主成分分析,通过保留信噪比较大的主成分,有效降低超光谱图像中的噪声水平;然后利用基于无监督正交子空间投影的端元提取算法获取图像中的端元矢量,将各端元矢量代入改进的约束能量最小化算子中,从而实现超光谱图像的目标检测。实验结果表明,该算法不但可以得到较好的目标检测结果,在运算性能上也具有一定的优势。A new target detection algorithm based on endmember extraction for use in hyperspectral image was proposed.On the condition of no prior knowledge,noise adjusting based principal component analysis was performed to the hyperspectral images.The noise level could be reduced by retaining several principal components with high signal-noise ratio.Then,endmember extraction method based on unsupervised orthogonal subspace projection was used to extract the endmember vectors,and the improved Constrained Energy Minimum（CEM） is implemented by using the endmember vectors for realizing the target detection of hyperspectral image.Experimental results show that the proposed algorithm has both better detection results and lower computational complexity.

关 键 词：目标检测 超光谱图像 端元提取 遥感 

分 类 号：V271.4[航空宇航科学与技术—飞行器设计] TP751.1[自动化与计算机技术—检测技术与自动化装置]