Incremental L1-Norm Linear Discriminant Analysis for Indoor Human Activity Classification
Incremental L1-Norm Linear Discriminant Analysis for Indoor Human Activity Classification
复制标题
DOI：
10.1109/radar.2019.8835593
复制
发表时间：
2019-04
期刊：
影响因子：
--
通讯作者：
Sivan Zlotnikov;Panos P. Markopoulos;F. Ahmad
中科院分区：
文献类型：
--
作者：
Sivan Zlotnikov;Panos P. Markopoulos;F. Ahmad
在本文中，我们提出了一种增量版本的l1 -范数线性判别分析（L1-LDA）用于基于雷达的室内人类活动分类。增量L1-LDA可以在操作过程中随着可用训练样本的增加而改进判别基。同时，它允许适应感兴趣的人类主体的特定活动模式，不同于原始判别基础训练的活动模式。增量版本保留了L1-LDA对训练数据中异常值的鲁棒性。利用各种室内人类活动的多普勒特征，我们证明了当训练数据损坏时，所提出的方法比标准线性判别分析的增量方法具有更高的性能，并且在标称训练数据下具有类似的性能。
In this paper, we present an incremental version of L1-norm Linear Discriminant Analysis (L1-LDA) for radar-based indoor human activity classification. Incremental L1-LDA enables refinement of the discriminant basis as more training samples become available during operation. At the same time, it permits adaptation to the specific activity patterns of the human subject of interest, different than the ones on which the original discriminant basis was trained. The incremental version retains the robustness of L1-LDA to outliers among the training data. Using Doppler signatures of various indoor human activities, we demonstrate that the proposed method exhibits enhanced performance over the incremental counterpart of standard linear discriminant analysis when the training data are corrupted and similar performance under nominal training data.