Balasingham Ilangko
Annu Int Conf IEEE Eng Med Biol Soc. 2016 Aug;2016:223-226. doi: 10.1109/EMBC.2016.7590680.
Improvement of classification performance is one of the key challenges in electroencephalogram (EEG) based motor imagery brain-computer interface (BCI). Recently, sparse representation based classification (SRC) method has been shown to provide satisfactory classification accuracy in motor imagery classification. In this paper, we aim to evaluate the performance of the SRC method in terms of not only its classification accuracy but also of its computation time. For this purpose, we investigate the performance of recently developed fast L1 minimization methods for their use in SRC, such as homotopy and fast iterative soft-thresholding algorithm (FISTA). From experimental analysis, we note that the SRC method with the fast L1 minimization algorithms is shown to provide robust classification performance, compared to support vector machine (SVM), both in time and accuracy.
提高分类性能是基于脑电图（EEG）的运动想象脑机接口（BCI）中的关键挑战之一。最近，基于稀疏表示的分类（SRC）方法已被证明在运动想象分类中能提供令人满意的分类准确率。在本文中，我们旨在不仅从分类准确率方面，而且从计算时间方面评估SRC方法的性能。为此，我们研究了最近开发的用于SRC的快速L1最小化方法的性能，如同伦方法和快速迭代软阈值算法（FISTA）。通过实验分析，我们注意到与支持向量机（SVM）相比，采用快速L1最小化算法的SRC方法在时间和准确率方面都表现出稳健的分类性能。
