Scholarlink

<p>Although brain-controlled typing based on imaginary handwriting typing has made significant breakthroughs, this research still faces some challenges, such as inaccurate character templates, easy confusion between similar characters, abnormal brain signal decoding, and erroneous detection of noise signals. To address these problems, the current study proposes the corresponding improvement approaches. Firstly, Sparse Principal Component Analysis (SPCA) is used to optimize character templates. Secondly, a second-level classifier based on Long Short-Term Memory (LSTM) network is designed to address the problem of misidentification of similar characters. Thirdly, the adaptive optimization algorithms are applied to the threshold setting and step size to mitigate the character recognition errors caused by abnormal attenuation of brain signal intensity and noise signals. Based on these improvements, the error rate of the characters decreased from 2.93% to 2.30%, the error rate of the words decreased from 14.39% to 11.21%, and the error rate of the sentences decreased from 73.02% to 60.32%. The study could offer important technical support for optimizing brain-controlled typing.</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

Research on optimization of brain-controlled typing recognition based on imaginary handwriting

  • 期刊论文

  • 发表日期：

    2025年10月17日

查看全文
  • Jiaofen Nan,
  • Panpan Xu,
  • Kaifan Zhang,
  • Qian Xu,
  • Siyuan Zhang,
  • Yongquan Xia,
  • Chuang Han,
  • Duan Li,
  • Yinghui Meng

摘要

Although brain-controlled typing based on imaginary handwriting typing has made significant breakthroughs, this research still faces some challenges, such as inaccurate character templates, easy confusion between similar characters, abnormal brain signal decoding, and erroneous detection of noise signals. To address these problems, the current study proposes the corresponding improvement approaches. Firstly, Sparse Principal Component Analysis (SPCA) is used to optimize character templates. Secondly, a second-level classifier based on Long Short-Term Memory (LSTM) network is designed to address the problem of misidentification of similar characters. Thirdly, the adaptive optimization algorithms are applied to the threshold setting and step size to mitigate the character recognition errors caused by abnormal attenuation of brain signal intensity and noise signals. Based on these improvements, the error rate of the characters decreased from 2.93% to 2.30%, the error rate of the words decreased from 14.39% to 11.21%, and the error rate of the sentences decreased from 73.02% to 60.32%. The study could offer important technical support for optimizing brain-controlled typing.