Scholarlink

<p>The classification of lower limb motor imagery (MI) is a significant challenge in brain-computer interface (BCI) research because of the closely analogous physiological representation of left and right lower limb movements in the human brain. A brain-computer interface (BCI) is a computer-based program that acquires, analyzes, and converts brain impulses into instructions transmitted to an output unit to perform the desired task. Motor imagery electroencephalogram (EEG) signals are produced by muscular activities. Most of the BCI system works on the information extracted from these signals. The EEG signals are usually contaminated by noise (commonly known as artifacts). Artifacts can appear due to undesirable actions including involuntary eye-blink, cardiac activities, and muscular movements. These artifacts contaminate the information, and hence need to be identified and removed before the processing of any EEG signals. In this article, we investigated to detect and eliminate ocular artifacts followed by classification of motor imagery EEG signals. We investigated various methods and compared the decomposition of EEG signals in the BCI-based motor imagery classification task. The experimental analysis presented in this work utilizes BCI competition III dataset IVA. The task is to accurately detect two classes (viz. right hand and right foot). Our results represents that the hybrid method involving independent component analysis &amp; recursive least squares (ICA-RLS) for ocular artifacts removal, Common spatial pattern (CSP) as well as other time and wavelet domain parameters are utilized for feature extraction, improved binary gravitation search algorithm (IBGSA) for effective electrode channel selection and combination of support vector machine classifier and particle swarm optimization (SVM-PSO) for classification, performed good results. The analytical results show that an average classification accuracy of 96.58% was obtained by the proposed model. It was also verified that the proposed approach performed better than other modern approaches for the same task.</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

Artificial intelligence-based performance enhancement in EEG signals classification for motor imagery system

  • 期刊论文

  • 发表日期：

    2025年8月7日

查看全文
  • Praveen Kumar Shukla,
  • Rahul Kumar Chaurasiya,
  • Vijay Kumar Sharma,
  • Amit Garg

摘要

The classification of lower limb motor imagery (MI) is a significant challenge in brain-computer interface (BCI) research because of the closely analogous physiological representation of left and right lower limb movements in the human brain. A brain-computer interface (BCI) is a computer-based program that acquires, analyzes, and converts brain impulses into instructions transmitted to an output unit to perform the desired task. Motor imagery electroencephalogram (EEG) signals are produced by muscular activities. Most of the BCI system works on the information extracted from these signals. The EEG signals are usually contaminated by noise (commonly known as artifacts). Artifacts can appear due to undesirable actions including involuntary eye-blink, cardiac activities, and muscular movements. These artifacts contaminate the information, and hence need to be identified and removed before the processing of any EEG signals. In this article, we investigated to detect and eliminate ocular artifacts followed by classification of motor imagery EEG signals. We investigated various methods and compared the decomposition of EEG signals in the BCI-based motor imagery classification task. The experimental analysis presented in this work utilizes BCI competition III dataset IVA. The task is to accurately detect two classes (viz. right hand and right foot). Our results represents that the hybrid method involving independent component analysis & recursive least squares (ICA-RLS) for ocular artifacts removal, Common spatial pattern (CSP) as well as other time and wavelet domain parameters are utilized for feature extraction, improved binary gravitation search algorithm (IBGSA) for effective electrode channel selection and combination of support vector machine classifier and particle swarm optimization (SVM-PSO) for classification, performed good results. The analytical results show that an average classification accuracy of 96.58% was obtained by the proposed model. It was also verified that the proposed approach performed better than other modern approaches for the same task.