基于空间分布特性多层RX高光谱点目标检测

Multi-layer RX Hyperspectral Point Target Detection Based on Spatial Distribution Characteristics

针对目前高光谱图像异常点目标检测过程中准确率低和虚警率高的问题,本文结合目标的空间分布特性,提出了一种多层级RX检测方法。通过计算被检测区域图像谱向相似性响应图,采用非线性抑制的方法,突出点目标并抑制背景。为进一步提高检测算法的表达能力和泛化性能,采用多级检测器级联的方式,逐层级增强异常点处的相对能量,削弱背景的影响,从而达到较高的检测性能。在外场挂飞试验数据集上进行验证,结果表明:该方法AUC值达到0.9881,明显优于CEM算法的0.9626和传统RX算法的0.9392。

In order to solve the problems of low accuracy and high false alarm rate in the detection process for abnormal point targets of hyperspectral images,a multi-level RX detection method based on the spatial distribution characteristics of targets is proposed.With the nonlinear suppression method,the point target can be highlighted and the background can be suppressed during the calculation for the spectral similarity response graph of the detected region.In order to further improve the expression ability and generalization performance of the detection algorithm,a cascade of multi-level detectors is used to enhance the relative energy of the outliers level by level and weaken the influence of the background so as to achieve high detection performance.The results show that the area under curve(AUC)value of this method reaches 0.9881,which is better than that of the constrained energy minimization(CEM)algorithm(0.9626)and the traditional RX algorithm(0.9392).