针对背景和目标的先验光谱特征未知的条件,给出一种基于单似然检验的高光谱图像小目标检测器。小目标相对于背景的低概率性使得高光谱图像数据对目标光谱信号的矩特征几乎不施加约束,可在最大熵条件下将广义似然比检验简化为对背景似然的单似然检验;利用全部数据样本建立无参估计模型以充分利用样本信息,从而得到基于单似然检验的高光谱图像小目标检测器。该检测器避免了统计模型误差和不明确物理含义特征对实际高光谱图像数据检测带来的影响。使用可见光/近红外波段机载I型实用型模块化成像光谱仪(OMIS-I)高光谱图像进行了实验,实验结果及相应理论分析表明该算法可有效检测高光谱图像中的空间低概率目标。

A small-target detector based on single likelihood test for hyperspectral imagery is presented to detect target when there is no a priori spectral signal of background and target, which presume the maximum entropy character of target. Because of the low-probability occurrence of target compared with that of background, it can be assumed that there is no constraint by hyperspectral imagery data on the moments of target signal. Accordingly, the generalize likelihood ratio test can be simplified to test background likelihood solely under maximum entropy of target. Then, nonparametric estimation is utilized to obtain the probability density of background, which can extract information from samples more effectively. The single likelihood test based detector weakens the effect of statistic model discrepancy and avoids effect of implicit physical meaning on detection. Theoretic analysis and the experimental results on visible/near-infrared OMIS-I hyperspectral imagery verify that these algorithms are effective to detect spatial low-probability targets.