一种改进的基于正交子空间投影的高光谱图像异常检测算法

Anomaly Detection Based on Improved Orthogonal Subspace Projection Algorithm in Hyperspectral Imagery

针对背景与目标能量差异较小的高光谱图像异常检测技术需求,提出了一种基于正交子空间投影的异常检测方法：首先利用正交子空间投影抑制背景信息;然后调整噪声,根据空间密度选取决策半径实现异常检测,并对结果进行形态滤波,消除大面积虚警。实验结果表明,本文算法能够检测到能量与背景差异较小的异常,且计算效率较高。

According to the demand of technology in hyperspectral anomaly detection with small energy differences between background and objectives, an improved orthogonal subspace projection algorithm is proposed. We firstly project the hyperspectral imagery onto the background orthogonal subspace to suppress background information. On this basis, after adjusting the noise, decision radiuscan be selected according to the spatial density realizing anomaly detection, and the morphological filter process as addition to eliminate the false alarm in large area. The experimental results show that, the algorithm proposed can detect the anomalies with small energy difference relative to the background, and has high calculation efficiency.