Scholarlink

<p>In this study, we present a novel FPGA-powered hybrid deep learning system for efficient drowsiness recognition based on electroencephalogram (EEG) signals. The proposed system integrates convolutional neural networks (CNNs) and long short-term memory (LSTM) networks, leveraging the strengths of each model to classify EEG signals effectively. To address the computational demands of hybrid models, we introduce a hardware accelerator architecture that enables parallel processing of both CNN and LSTM sub-models. Employing eight processing elements per layer and optimizing memory usage through array partitioning, unrolling, and memory reuse, our system significantly enhances processing speed while maintaining high accuracy. Experimental results demonstrate the superiority of our approach, achieving an impressive accuracy of 77.76% and an inference time of 0.073&#xa0;s per EEG signal classification, outperforming state-of-the-art methods. Additionally, we propose optimizations such as using hard sigmoid and hard tanh activation functions and implementing SVD-based low-rank approximation to further improve performance. Our FPGA-accelerated hybrid model offers a promising solution for rapid and accurate drowsiness detection, with implications for improving road safety. GitHub link: <a href="https://github.com/MUNI9849/FPGA-design-for-hybrid-CNN-LSTM-model">https://github.com/MUNI9849/FPGA-design-for-hybrid-CNN-LSTM-model</a>.</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

FPGA-accelerated hybrid CNN-LSTM system for efficient EEG-based drowsiness recognition

  • 期刊论文

  • 发表日期：

    2025年1月31日

查看全文
  • Rama Muni Reddy Yanamala,
  • Muralidhar Pullakandam

摘要

In this study, we present a novel FPGA-powered hybrid deep learning system for efficient drowsiness recognition based on electroencephalogram (EEG) signals. The proposed system integrates convolutional neural networks (CNNs) and long short-term memory (LSTM) networks, leveraging the strengths of each model to classify EEG signals effectively. To address the computational demands of hybrid models, we introduce a hardware accelerator architecture that enables parallel processing of both CNN and LSTM sub-models. Employing eight processing elements per layer and optimizing memory usage through array partitioning, unrolling, and memory reuse, our system significantly enhances processing speed while maintaining high accuracy. Experimental results demonstrate the superiority of our approach, achieving an impressive accuracy of 77.76% and an inference time of 0.073 s per EEG signal classification, outperforming state-of-the-art methods. Additionally, we propose optimizations such as using hard sigmoid and hard tanh activation functions and implementing SVD-based low-rank approximation to further improve performance. Our FPGA-accelerated hybrid model offers a promising solution for rapid and accurate drowsiness detection, with implications for improving road safety. GitHub link: https://github.com/MUNI9849/FPGA-design-for-hybrid-CNN-LSTM-model.