结合KI准则和逆高斯模型的SAR影像非监督变化检测

A Novel Approach Combining KI Criterion and Inverse Gaussian Model to Unsupervised Change Detection in SAR Images

摘要: 提出一种结合逆高斯模型（inverse Gaussian model，IGM）和KI（Kittler-Illingworth，KI）最小错误率准则的合成孔径雷达（synthetic aperture radar，SAR）影像非监督变化检测方法。假设差值影像中未变化类和变化类服从混合IGM，结合贝叶斯决策理论，自动求取满足KI最小错误率准则的阈值。在两组多时相SAR数据上分别设计了两组实验以验证本文方法的有效性。实验表明，本文方法可以更好地估计差值影像中未变化类和变化类的概率密度分布，得到合理的决策阈值，有效提高变化检测图的精度。

Abstract: In this context, a novel approach combining inverse Gaussian model (IGM) and the Kittler-Illingworth (KI) criterion has been proposed to carry out tunsupervised change detection in synthetic aperture radar (SAR) images. The minimum error threshold could be computed by exploiting the Bayes decision theory under the assumption that hybrid IGM could describe the distribution of the changed and unchanged class in difference image. Experiments carried out on two sets of multi-temporal SAR images indicate that the proposed approach can effectively estimate the probability density function of the unchanged and changed classes in the difference image and acquire a reasonable threshold for yielding a better change map from the difference image.