光学学报, 2016, 36 (1): 0130003, 网络出版: 2015-12-31
基于滑窗负熵统计的高光谱独立特征提取方法
Independent Component Feature Extraction Method for Hyperspectral Image Based on Negentropy Statistics in Moving Windows
光谱学 特征提取 光谱降维 滑窗统计 独立成分分析 目标检测 spectroscopy feature extraction spectral dimension reduction moving window statistics independent component analysis target detection
摘要
针对独立成分的排序问题,提出一种滑窗负熵统计的独立成分特征提取方法,并将优选成分应用于目标检测。采用小窗口在独立成分的二维空间范围滑动,采用非多项式负熵逼近统计评价每个窗口数据,将窗口统计值的最大值作为该成分的评价值,按照评价值大小排序;采用直方图零值分割方法,确定有效独立成分阈值,将独立成分进行二值化,实现特征提取后独立成分的目标检测。实验结果表明,基于滑窗负熵统计的独立成分特征提取方法,既能避免野值对评价优选结果的影响,又能优选出含有较小目标的独立成分,便于感兴趣目标的快速检测。
Abstract
An independent component feature extraction method based on negentropy statistics in moving windows is presented for the ordering of the independent components, and applied in target detection. A small window is moved in the two dimensional space of the independent component image. Data from each window is evaluated by negentropy approximation statistics using nonpolynomial function. The largest one of all of the evaluations is considered as the result evaluation, and the component images are ordered by the result evaluation. The two- value figure from the chosen component is made by histogram zero value split method, realizing target detection from the feature extracted independent components. The experiment results show that the independent component feature extraction method based on negentropy statistics in moving windows can avoid the influence of wild values, also select the valid components with small target, and benefit rapid detection of interested target.
朱院院, 高教波, 高泽东. 基于滑窗负熵统计的高光谱独立特征提取方法[J]. 光学学报, 2016, 36(1): 0130003. Zhu Yuanyuan, Gao Jiaobo, Gao Zedong. Independent Component Feature Extraction Method for Hyperspectral Image Based on Negentropy Statistics in Moving Windows[J]. Acta Optica Sinica, 2016, 36(1): 0130003.