摘要： 针对多时相高光谱影像的微弱变化,提出一种基于最小平均绝对误差(MMAE)的无监督变化信息提取方法。利用高光谱影像光谱特征之间的内在联系,对光谱特征向量组成的时间序列进行单波段向量自回归模型预测,得到局部变化趋势以及进行多波段拟合获得整体变化趋势,并通过哈达玛积将两者相结合,利用MMAE指数有效地提取微弱变化信息,获得初步变化信息图。实验结果表明,与SCD,PCA,CVA,IR-MAD等方法相比,该方法能够更有效地提取多时相高光谱影像的微弱变化信息,保持变化区域的细节,同时可抑制不同时相高光谱影像的背景噪声。

关键词: 微弱变化, 高光谱影像, 最小平均绝对误差, 向量自回归, 变化信息提取, 变化检测

Abstract: This paper proposes a novel unsupervised method for slight change information extraction from multitemporal hyperspectral image sequence based on the Minimum Mean Absolute Error(MMAE) index.To consider the spectral signatures in hyperspectral data,the Vector Autoregressive(VAR) and fitting models are exploited to create a prediction of single-band and multi-bands time series,while MMAE index is introduced to generate the Preliminary Change Information Image(PCII),which contains significant change information.Experimental results show that the proposed method is more effective in extracting the slight change information effectively in the hyperspectral image sequence preserving significant change details and suppressing random variation of the background.compared with the SCD,PCA,CVA and IR-MAD methods.

Key words: slight change, hyperspectral image, Minimum Mean Absolute Error(MMAE), Vector Autoregressive(VAR), change information extraction, change detection

中图分类号: 

• TP391