基于三维GMRF的多光谱图像自适应目标检测

The Adaptive Target Detection Based on 3-Dimensional Gauss Markov Random Field Model in Multispectral Imagery

针对常见的多光谱目标检测算法仅利用光谱信息的局限性,提出一种移动窗口局部异常自适应检测方法。采用加性目标信号和非结构化背景模型描述多光谱图像数据;基于谱间相关性和空间相关性,利用三维高斯马尔可夫随机场(GMRF)模型估计背景数据二阶统计量的逆,最后通过广义似然比检验实现了自适应目标检测。仿真试验及其理论分析表明了算法的有效性。

In this paper, an automatic target detection algorithm for multispectral imagery based on pixel-moving widow is proposed to overcome the limitations of traditional target detection algorithms which utilize only the spectral information while discarding the useful spatial information. A target plus unstructured background addictive model is constructed; Then, global second order statistics of background is estimated via 3-dimensional Gauss Markov random field （GMRF） avoiding the difficulty in computing inverse; finally, within each local data cube split by the moving window; detection is accomplished by generalized likelihood ratio test comparing with the classical RXD, experimental results verify the effectiveness of our algorithm.