基于增量学习的高光谱图像目标检测

高光谱图像目标检测是高光谱图像分析中的重要研究内容之一。本文从经典有效的约束能量最小化算法出发,提出了一种基于增量学习的高光谱目标检测方法。当获得新的样本时,不需要重新计算所有样本的自相关矩阵即可对检测器模型进行更新,减轻了星上有限计算资源的负担。实验结果表明:本文提出的目标检测算法在压制背景光谱的同时可以更好地适应目标光谱,提高了算法的检测性能。

Hyperspectral Target Detection Algorithm Based on Incremental Learning

Hyperspectral image target detection is one of the most important research directions in hyperspectral image analysis field. In this paper, an incremental learning based target detection method is proposed. We design the incremental form of the classical constrained energy minimization algorithm. When a new sample is obtained, the proposed method will update the model without recalculating the auto-correlation matrix of all samples, which reduces the burden of on-board calculations. Experiments show that the proposed method can adapt the target spectrum while suppressing the background spectrum, which improves the detection accuracy effectively.