为了克服基于线性混合模型的高光谱遥感影像亚像元目标探测方法的缺陷, 提出了一种基于全限制性线性分解的自适应匹配子空间探测方法。首先利用交叉相关光谱匹配技术求得各个像元所含端元类别信息, 然后根据端元类别信息和全限制性分解的结果构造自适应匹配子空间探测算子, 利用端元类别信息在探测中动态选择端元, 降低端元数目估计偏差对探测结果的影响, 提高探测器对目标与背景的可分性。实验证明, 该方法与其他基于线性混合模型的亚像元目标探测方法相比, 可以更好地克服端元数目估计偏差对探测结果的影响, 无论是端元个数低估还是高估时,探测效果均更优。

This paper presents an adaptive matched subspace method for detecting sub-pixel targets in hyperspectral imagery based on fully constrained linear separation. This method aims to overcome the defects of the sub-pixel detecting methods based on linear mixture model. By means of this method, not only the abundance of targets in different pixels can be detected, but also the pixels containing targets can be separated from the other pixels reliably. In addition, cross correlation spectrum matching technique is applied to the method to compute the sorts of the endmembers in each pixel in the imagery. Then instead of choosing all the endmembers, we choose the according sorts of endmembers in the method. In this way, the separability between the targets and the other ground objects can be improved. The experiments show that no matter whether the number of the sorts of endmembers is overestimated or underestimated, the detecting results of the method presented in this paper are better than other traditional sub-pixel detecting methods based on linear-mixture model. And this method can formulate an effective rule to separate the targets and background with a better performance than the other methods. Besides, it also performs better as to the targets spectrally similar to the background objects and the targets with a small number.