Electroencephalogram (EEG) is a recording of comprehensive reflection of physiological brain activities. Because of many reasons, however, including noises of heartbeat artifacts and muscular movements, there are complex challenges for efficient EEG signal classification. The Convolutional Neural Networks (CNN) is considered a promising tool for extracting data features. A deep neural network can detect the deeper-level features with a multilayer through nonlinear mapping. However, there are few viable deep learning algorithms applied to BCI systems. This study proposes a more effective acquisition and processing HW-SW method for EEG biosignal. First, we use a consumer-grade EEG acquisition device to record EEG signals. Short-time Fourier transform (STFT) and Continuous Wavelet Transform (CWT) methods will be used for data preprocessing. Compared with other algorithms, the CWT-CNN algorithm shows a better classification accuracy. The research result shows that the best classification accuracy of the CWT-CNN algorithm is 91.65%. On the other side, CNN inference requires many convolution operations. We further propose a lightweight CNN inference hardware accelerator framework to speed up inference calculation, and we verify and evaluate its performance. The proposed framework performs network tasks quickly and precisely while using less logical resources on the PYNQ-Z2 FPGA development board.
In this article, we provide a brief overview of the EEG-based classification of motor imagery activities using machine learning methods. We examined the effect of data segmentation and different neural network structures. By applying proper window size and using a purely convolutional neural network, we achieved 97.7% recognition accuracy on data from twenty subjects in three classes. The proposed architecture outperforms several networks used in previous research and makes the motor imagery-based BCI more efficient in some applications. In addition, we examined the performance of the neural network on a FPGA-based card and compared it with the inference speed and accuracy provided by a general-purpose processor.
1 December 2019 in Carpathian Journal of Electronic and Computer Engineering
Machine-learning techniques allow to extract information from electroencephalographic (EEG) recordings of brain activity. By processing the measurement results of a publicly available EEG dataset, we were able to obtain information that could be used to train a feedforward neural network to classify two types of volunteer activities with high efficiency.
19 May 2021 in Annales Mathematicae et Informaticae