Scholarlink

<p>A brain–computer interface (BCI) enables the device to communicate directly with the brain by decoding neural signals, particularly electroencephalograms (EEGs). EEG signals are used in a variety of applications, especially motor imagery detection, due to their noninvasive nature, real-time monitoring capabilities, and cost-effectiveness. Often, EEG data consists of multi-channel signals; the presence of multiple channels leads to computational complexity and the presence of redundant channel signals. To avoid these, the channel selection algorithm is currently being used, particularly optimization-based channel selection. However, the optimization-based channel selection method may have limitations, such as eliminating the most important channel due to poor initialization and failing to achieve optimal performance due to the lack of an efficient multi-objective fitness function. To address these limitations, we proposed a new channel selection mechanism called the multi-objective prioritized jellyfish search algorithm (MPJS), which has two significant improvements. First, domain-specific initialization is employed to select the most important channels at the initialization stage, which ensures that no important channels are omitted. Second, using a multi-objective fitness function instead of a single objective one to select the most relevant and informative channels ensures that the selected channels meet the number criteria and include candidates’ channels. Prior work primarily focused on two-class MI classification, with only a few studies examining four-class MI classification; however, these four-class classification methods fail to achieve optimal performance. To address these research gaps and achieve optimal performance in four-class MI detection, we proposed an improved double-branch EEGNET (DB-EEGNET). This proposed&#xa0;work performance was evaluated by using benchmark datasets, including BCI Competition IV-2008-2A, BCI Competition III-2008-A, and the High Gamma dataset (HGD). Our proposed MJPS channel selection and DB-EEGNET classification method outperformed the baseline algorithm on the BCI IV-IIA, IIIA, and HGD datasets, with an average accuracy of 83.9%, 84.46%, and 94.78%, respectively.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Automatic channel selection using multi-objective prioritized jellyfish search (MPJS) algorithm for motor imagery classification using modified DB-EEGNET

  • 期刊论文

  • 发表日期：

    2025年1月23日

查看全文
  • D. Senthil Vadivelan,
  • Prabhu Sethuramalingam

摘要

A brain–computer interface (BCI) enables the device to communicate directly with the brain by decoding neural signals, particularly electroencephalograms (EEGs). EEG signals are used in a variety of applications, especially motor imagery detection, due to their noninvasive nature, real-time monitoring capabilities, and cost-effectiveness. Often, EEG data consists of multi-channel signals; the presence of multiple channels leads to computational complexity and the presence of redundant channel signals. To avoid these, the channel selection algorithm is currently being used, particularly optimization-based channel selection. However, the optimization-based channel selection method may have limitations, such as eliminating the most important channel due to poor initialization and failing to achieve optimal performance due to the lack of an efficient multi-objective fitness function. To address these limitations, we proposed a new channel selection mechanism called the multi-objective prioritized jellyfish search algorithm (MPJS), which has two significant improvements. First, domain-specific initialization is employed to select the most important channels at the initialization stage, which ensures that no important channels are omitted. Second, using a multi-objective fitness function instead of a single objective one to select the most relevant and informative channels ensures that the selected channels meet the number criteria and include candidates’ channels. Prior work primarily focused on two-class MI classification, with only a few studies examining four-class MI classification; however, these four-class classification methods fail to achieve optimal performance. To address these research gaps and achieve optimal performance in four-class MI detection, we proposed an improved double-branch EEGNET (DB-EEGNET). This proposed work performance was evaluated by using benchmark datasets, including BCI Competition IV-2008-2A, BCI Competition III-2008-A, and the High Gamma dataset (HGD). Our proposed MJPS channel selection and DB-EEGNET classification method outperformed the baseline algorithm on the BCI IV-IIA, IIIA, and HGD datasets, with an average accuracy of 83.9%, 84.46%, and 94.78%, respectively.