基于扩展数学形态学的高光谱图像异常检测

提出了一种新型的基于扩展数学形态和光谱相似度测量的高光谱图像异常榆测方法.在日标与背景未知的情况下.同时利用光谱和空间信息实现日标的定位与检测,实现离光谱遥感数据的日标检测.通过扩展的膨胀和腐蚀操作实现目标特征提取;通过正交投影散度计算扩展形态学操作的累加距离确定排序关系并利用其融合特征提取结果实现特征提取结果的融合.算法性能通过合成的OMIS数据进行评价.与经典异常检测RX算法进行比较.并应用于具有相似光谱特征目标的区分.实验证明,本文提出的算法性能优于RX算法.具有低虚警率的异常目标检测结果,并且能够较好地区分了相似光谱特征的异常日标.

Anomaly Detection Based on Extended Mathematical Morphology for Hyperspectral Imagery

A novel anomaly detection algorithm based on the theory of extended mathematical morphology and spectral similarity measurement is proposed for hyperspectral imagery. The spatial and spectral information has been used to locate and detect targets under the condition of none prior knowledge of targets and background. The extended mathematical morphological erosion and dilation operations are performed respectively to extract the targets features. The orthogonal projection divergence is used to calculate the cumulative distance in the erosion and dilation operations to determine the ordering relation. And the orthogonal projection divergence is also performed to measure the spectral similarity to fuse the results of feature extraction. The synthesized hyperspectral images collected by object modularization imaging spectrometer (OMIS) is applied to evaluate the proposed algorithm, the proposed algorithm is compared with RX algorithm by a specifically designed experiment, andit is applied to distinguish the targets with similar spectral characteristics. From the results of experiments, it is illuminated that the proposed algorithm can detect anomalous targets with low false alarm rate and its performance is better than that of RX algorithm under the same condition. It is also illuminated that the proposed algorithm can differentiate targets with similar spectral characteristics well with low false alarm rate.