提出了一种基于局域信息的核化正交子空间投影的目标探测方法(KLOSP)。模拟数据实验证明,KLOSP方法比其他子空间目标探测方法具有更优的接受者操纵特征曲线;真实影像数据实验证明,该方法比传统子空间目标探测方法具有更大的目标与背景的可分度,能够准确地对高光谱影像数据进行目标探测。
In order to optimize the background subspace and suppress the false alarm better, this paper proposes a local information-based kernelized OSP method (KLOSP) for target detection. Neighbor spatial information is brought in to construct variable optimum background projective subspace. KLOSP has acquired the best Receiver Operating Characteristics (ROC) curve in simulated data experiment and obtained the biggest target-background difference as well as the highest detection rate. It is proved that this algorithm can detect targets in hyperspectral imageries accurately and effectively in real image experiment.