Scholarlink

<p>Electroencephalography (EEG) is a noninvasive technique used to record brain electrical activity. It has been widely explored for brain–computer interface applications, including imagined speech analysis. However, handling and utilizing large EEG datasets have challenges due to their size. We utilize Siamese neural networks (SNNs) to improve data utilization through pairwise learning by creating multiple training combinations of EEG data. SNNs are well-suited for tasks that involve comparing two or more inputs, making them ideal for analyzing EEG signals. This research focuses on utilizing SNNs to identify different brain states based on EEG signal analysis. A novel publicly available collection of electroencephalogram (EEG) recordings extracted from the imaginary pronunciation of two sets of Spanish words by 15 healthy people is presented. The first set includes 5 Spanish vowels, while the second set represents the instructions up, down, front, back, right and left. EEG signals were captured at 1024 Hz using a six-channel acquisition device. Each word was repeated fifty times by every randomly chosen participant. For contrast, certain blocks were recorded under the "pronounced speech" setting, which involves simultaneously acquiring audio and EEG signals. Subsequently, a DWT-based preprocessing and autoencoder-assisted Siamese Neural Network with triplet loss is proposed for EEG signal analysis. The results indicate that different brain states can be accurately identified with a high degree of accuracy using the proposed methodology.</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

Analyzing imagery speech and pronounced speech using Siamese neural networks (SNNs) for brain human machine interface (BHMI) classification

  • 期刊论文

  • 发表日期：

    2026年3月28日

查看全文
  • Sachin Kumar Shaw,
  • Subhranil Som,
  • Amlan Chakrabarti,
  • Ankit Garg,
  • Rabi Shaw

摘要

Electroencephalography (EEG) is a noninvasive technique used to record brain electrical activity. It has been widely explored for brain–computer interface applications, including imagined speech analysis. However, handling and utilizing large EEG datasets have challenges due to their size. We utilize Siamese neural networks (SNNs) to improve data utilization through pairwise learning by creating multiple training combinations of EEG data. SNNs are well-suited for tasks that involve comparing two or more inputs, making them ideal for analyzing EEG signals. This research focuses on utilizing SNNs to identify different brain states based on EEG signal analysis. A novel publicly available collection of electroencephalogram (EEG) recordings extracted from the imaginary pronunciation of two sets of Spanish words by 15 healthy people is presented. The first set includes 5 Spanish vowels, while the second set represents the instructions up, down, front, back, right and left. EEG signals were captured at 1024 Hz using a six-channel acquisition device. Each word was repeated fifty times by every randomly chosen participant. For contrast, certain blocks were recorded under the "pronounced speech" setting, which involves simultaneously acquiring audio and EEG signals. Subsequently, a DWT-based preprocessing and autoencoder-assisted Siamese Neural Network with triplet loss is proposed for EEG signal analysis. The results indicate that different brain states can be accurately identified with a high degree of accuracy using the proposed methodology.