基于多孔径映射的高光谱异常检测算法

针对高光谱遥感异常检测中复杂背景与异常目标之间光谱特征相关性导致背景模型难以准确估计的问题,提出了一种基于多孔径映射的高光谱遥感异常检测算法。首先,不同于背景建模提取背景特征的方法,多孔径映射从不同角度提取数据特征,通过构建基集合表征高光谱数据的光谱特性,获得用于衡量统计差异的异常显著性指标。其次,为了实现对具有适中及低异常显著性像素的精细分析,本文基于模糊逻辑理论构建隶属度函数获得关于像素异常显著性的连续性属性标记,并将隶属度值作为权重,通过加权迭代过程实现多孔径映射的自适应收敛。最后,借鉴模糊逻辑理论中的去模糊机制,对多孔径检测结果进行融合,获得最终的检测结果。本文仿真试验采用高光谱遥感数据,从稳健性及对低显著度目标敏感性方面对算法进行验证。

An Anomaly Detector Based on Multi-aperture Mapping for Hyperspectral Data

Considering the correlationship of spectral content between anomaly and clutter background, inaccurate selection of background pixels induced estimation error of background model.In order to solve the above problems,a multi-aperture mapping based anomaly detector was proposed in this paper.Firstly, differing from background model which focused on feature extraction of background,multi-aperture mapping of hyperspectral data characterized the feature of whole hyperspectral data.According to constructed basis set of multi-aperture mapping,anomaly salience index of every test pixel was proposed to measure the relative statistic difference.Secondly,in order to analysis the moderate salience anomaly precisely,membership value was constructed to identify anomaly salience of test pixels continuously based on fuzzy logical theory.At same time,weighted iterative estimation of multi-aperture mapping was expected to converge adaptively with membership value as weight.Thirdly,classical defuzzification was proposed to fuse different detection results.Hyperspectral data was used in the experiments,and the robustness and sensitivity to anomaly with lower silence of proposed detector were tested.