董燕妮, 杜博, 张乐飞, 张良培. 基于局域自适应信息理论测度学习的高光谱目标探测方法[J]. 武汉大学学报 ( 信息科学版), 2018, 43(8): 1271-1277. DOI: 10.13203/j.whugis20150504
引用本文: 董燕妮, 杜博, 张乐飞, 张良培. 基于局域自适应信息理论测度学习的高光谱目标探测方法[J]. 武汉大学学报 ( 信息科学版), 2018, 43(8): 1271-1277. DOI: 10.13203/j.whugis20150504
DONG Yanni, DU Bo, ZHANG Lefei, ZHANG Liangpei. Hyperspectral Target Detection Based on Locally Adaptive Information-Theoretic Metric Learning Method[J]. Geomatics and Information Science of Wuhan University, 2018, 43(8): 1271-1277. DOI: 10.13203/j.whugis20150504
Citation: DONG Yanni, DU Bo, ZHANG Lefei, ZHANG Liangpei. Hyperspectral Target Detection Based on Locally Adaptive Information-Theoretic Metric Learning Method[J]. Geomatics and Information Science of Wuhan University, 2018, 43(8): 1271-1277. DOI: 10.13203/j.whugis20150504

基于局域自适应信息理论测度学习的高光谱目标探测方法

Hyperspectral Target Detection Based on Locally Adaptive Information-Theoretic Metric Learning Method

  • 摘要: 传统基于信号检测的目标探测方法需要依赖特定的统计假设,只有在符合条件的情况下才能取得较好的目标探测结果。为了克服这一缺陷,提出了一种基于局域自适应的信息理论测度学习方法。首先将信息理论测度学习方法作为目标主函数,然后加以局域自适应决策法则进行约束,自适应地减小相似样本对距离,增大不相似样本对距离,使得在考虑阈值的同时兼顾测度学习前后距离的改变情况来进行目标探测决策,从而更好地实现目标探测。实验证明,该方法与其他经典目标探测方法或测度学习方法相比,可以更好地实现目标与背景分离,能够更有效地对高光谱影像数据进行目标探测。

     

    Abstract: The classical target detection methods may only perform well with certain assumptions. To overcome the shortcomings, this paper proposes a novel local decision adaptive information-theoretic metric learning (LA-ITML) target detector. Firstly, the proposed method uses the ITML method as the objective function for learning a Mahalanobis distance to separate similar and dissimilar point-pairs. Then, a locally decision adaptive constraint is applied to shrink the distances between samples of similar pairs and expand the distances between samples of dissimilar pairs. Finally, we can make the detection decision by considering both the threshold and the changes between the distances before and after metric learning. The experimental results demonstrate that the proposed method can obviously separate target samples from background ones and out performs both the state-of-the-art target detection algorithms and the other classical metric learning methods.

     

/

返回文章
返回