利用最小噪声分数变换的高光谱目标探测方法

Noise Fraction Minimization Based Target Detection Algorithm for Hyper-spectral Imagery

针对高光谱遥感影像特征维数多、冗余度高,同时观测信号中广泛存在各种噪声成分的问题,该文提出了一种利用特征降维的高光谱遥感影像目标探测方法。方法首先采用最小噪声分数变换算法对高光谱遥感数据进行特征降维,提取观测信号中的少量高信噪比成分组成新的数据立方,进而使用高光谱目标探测算子实现目标探测。通过对真实目标位置已知的HyMap高光谱遥感数据目标探测的实验,证明了该文提出的方法能够降低目标探测的虚警率和探测时间。

In order to address the issues of high feature dimension,the redundancy among the cross-channels,and the computational cost,a method was proposed in this paper to detect the sub-pixel targets from hyper-spectral images based on feature dimension reduction.In this paper,the noise fraction minimization transform is introduced for spectral dimension reduction,which extracted the principal components in accordance with the signal to noise ratio(SNR)and formed the new data cube for subsequent target detection.Experiments on the famous HyMap data set named'Target Detection Self-Test'indicated that proposed approach could reduce the false alarm rate and processing time.