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.
出处
《电光与控制》
北大核心
2010年第8期45-48,共4页
Electronics Optics & Control
关键词
目标检测
超光谱图像
端元提取
遥感
target detection
hyperspectral image
endmember extraction
remote sensing
