Parallel convolutional neural network and empirical mode decomposition for high accuracy in motor imagery EEG signal classification.

Motor imagery EEG (MI-EEG) signal classification is one of the key challenges in Brain-computer interface (BCI) technology. Currently, MI-EEG signal analysis methods based on Riemannian geometry are usually based on a single symmetric positive definite (SPD) manifold. This approach offers a relatively simple geometric structure. However, it struggles to fully capture the complex geometric characteristics of MI-EEG signals. To address this, this paper introduces the Gaussian SPD manifold in addition to the classical SPD manifold. The method extracts more discriminative kernel features from both manifolds. These features are then fused for improved classification performance. The Gaussian SPD manifold models the traditional SPD manifold with a Gaussian distribution. By doing so, it can more accurately describe the statistical characteristics and variations of MI-EEG signals, compensating for the limitations of the single SPD manifold. The method was validated on the OpenBMI binary classification dataset, achieving an average classification accuracy of $80.8 \%$ with a standard deviation of $10.8 \%$. The approach presented in this paper provides a new perspective for the geometric analysis of MI-EEG signals, allowing for a deeper understanding and analysis of their complex geometric structure.

By 2021, WHO projects over a billion incapacitated people, with 20% facing daily functional impairments.Brain-Computer Interface (BCI) offers effortless machine control via direct brain-computer interaction, with Motor Imagery (MI) Electroencephalogram (EEG) as the key BCI foundation.The MI EEG signals were collected from the BNCI horizon2020 database for nine participants.The MI EEG data includes four tasks: imaginative movement of the left hand, right hand, feet, and tongue.There are 22 EEG channels with a sampling rate of 250Hz in the data.The MI EEG signals were band-pass filtered with a lower cut-off frequency of 0.5 Hz and an upper cut-off frequency of 100 Hz.Based on the energy count threshold approach (ECTA) nine channels were identified as dominating channels from the filtered MI EEG signals.Energy values for each channel were extracted in the ECTA method for the 3-sec window.And if the energy value of a channel for a particular window is greater than 60% of the maximum channel's energy, then the energy count value will be incremented by one.Finally whichever channels had larger energy counts were identified as a dominant channel.On these nine dominant MI EEG signals discrete wavelet transform using Daubechies 4 mother wavelet a four-level decomposition is applied, and Mu and beta rhythms were extracted.For a 3-sec window from the nine dominant channels, the energy and entropy feature values from the Mu and Beta rhythms were extracted.From the extracted features, 80% of the data is used for training Linear Discriminant Analysis (LDA) and Support Vector Machine (SVM) and 20% for testing classifier models.Both the models performed well on the test data and the results obtained had a highest accuracy 91.72.7 for subject-9 and these results were compared with the existing methods.The obtained results of MI EEG signals classification using BCI holds potential for revolutionizing assistive technology, stroke rehabilitation, virtual reality gaming,

A convolutional spiking neural network with adaptive coding for motor imagery classification

A classification method for EEG motor imagery signals based on parallel convolutional neural network

Background: Motor-Imagery (MI) is a mental or cognitive stimulation without actual sensory input that enables the mind to represent perceptual information. This study aims to use the optimized support vector machine (OSVM) by differential evolution algorithm for motor-Imagery EEG signal classification. Methods: A total of three filters were applied to each signal during the preprocessing phase. The bandstop filter was used to remove urban noise and signal recorders, the median filter to remove random sudden peaks in the signal, and finally, the signal was normalized using the mapminmax filter. The most valuable features were extracted including mean signal intensity, minimum signal value, signal peak value, signal median, signal standard deviation, energy, corticoids, entropy, and signal skewness. Results: The accuracy of the SVM for linear, Gaussian, polynomial, and radial base kernels was 67.3%, 55.1%, 63.6%, and 55.1%, respectively, which was optimized after the classification model by differential evolution algorithm; however, the accuracy for OSVM was increased to 99.6%. Conclusion: Examination of the brain signal appearance for uniform motor-Imagery of both hands showed a significant difference between the signal of motor-Imagery mode with OSVM algorithm (99.6% accuracy), which gave promising results for classification motor imagery EEG signal.

Classification of electroencephalography (EEG) signals for brain-computer interface has great impact on people having various kinds of physical disabilities. Motor imagery EEG signals of hand and leg movement classification can help people whose limbs are replaced by prosthetics. In this paper, random subspace ensemble network with variable length feature sampling has been proposed for improving the prediction accuracy of motor imagery EEG signal classification. The method has been tested on eight different subjects and a hybrid dataset of two subjects data combined. Discrete wavelet transform based de-noising scheme has been adopted to remove artifacts from the EEG signal. For sub-band selection, dual-tree complex wavelet Transform has been employed. Mutual information scoring has been used for univariate feature selection from the feature space. A comparative analysis has been carried out where random subspace ensemble network outperformed other classification models. The maximum accuracy obtained by the model was 90.00%. Furthermore, the model showed better performance on the hybrid dataset with an average accuracy of 86.00%. The findings of this study are expected to be useful in artificial limb movements through brain-computer interfacing for rehabilitation of people with such physical disabilities.

Chapter 8 - Ensemble Learning Approach to Motor Imagery EEG Signal Classification

Motor imagery EEG recognition based on conditional optimization empirical mode decomposition and multi-scale convolutional neural network

The electroencephalogram (EEG) signal from motor imagery (MI) is used to drive brain–computer interaction (BCI). However, users usually are not adept at performing MI, which leads to low-quality EEG signals and decreases the performance of BCI applications. The humanoid robot stimulation approach can guide users in performing MI more proficiently by increasing the cortico-spinal excitability and improving the discrimination of event-related desynchronization patterns during MI tasks. Compared to the traditional stimulation modes, our proposed humanoid robot stimulation mode can activate higher quality MI EEG signals. We use convolutional neural network and long short-term memory algorithm for the extraction of EEG features and classification. The results showed that the CNN-LSTM can achieve the highest classification accuracy (93.7% ±1.7%) in the humanoid robot stimulation mode, and it outperformed all other classifier-stimulation mode combinations. This demonstrates the effectiveness and feasibility of using a humanoid robot in real-scene MI-BCI applications, such as service robots or rehabilitation system for person with motor disabilities.

A motor imagery EEG signal classification algorithm based on recurrence plot convolution neural network

One of the essential challenges in brain–computer interface is to classify motor imagery (MI) signals. In this paper, an ensemble SVM-based voting system is proposed. In each line of this system, the EEG signal is transformed into different representations based on discrete cosine transform, Fourier transform, common spatial pattern, and empirical mode decomposition, and then these representations are combined in a triple-frame matrix. These frames are fed into a pre-trained deep convolutional neural network as a feature extractor. For each line, an SVM is employed to classify the extracted feature vectors. Finally, a decision is made based on voting between these SVMs. Performance of the proposed method is examined on the BCI Competition III dataset Iva to separate right hand and foot movement imagery. The simple proposed method achieves the average accuracy of 96.34% for all of the subjects, and 99.70% for the best situation that is an improvement in MI classification. In addition, it can be seen that right side of the brain is more effective than the other side in EEG-based MI classification.

In recent years, with the development of social economy and technology, the brain-computer interface based on motor imagery(MI-BCI) has gradually become the focus content of many re-searchers. However, the motor imagery EEG signal (MI-EEG) itself has the characteristics of non-linearity and low signal-to-noise ratio, and because the characteristics of different domains of MI-EEG cannot be effectively combined, the recognition rate of MI-EEG is unsatisfactory. To overcome the above problems, this paper proposes a Transformer-based one-dimensional convolutional neural network model(CNN-Transformer) for the classification and recognition of four types of motor imagery EEG signals. Firstly, the artifacts of the original EEG are removed and new time-space-frequency features are constructed by preprocessing such as bandpass filtering and PCA dimensionality; then, the local features in the temporal dimension are extracted through the convolution and pooling operations of 1D-CNN, while reducing the dimension of the time feature; next, the Transformer based on the attention mechanism is used to extract more abstract and high-level temporal features from multiple perspectives; finally, the classification results are integrated and output through the fully connected layer. The performance of the CNN-Transformer model is evaluated using the competition dataset 2008 BCI-Competition 2A. The results show that the average accuracy and kappa value of the CNN-Transformer model are as high as 99.29%(±0.07%) and 98.43%(±0.21), respectively, which are 3.72% and 7.68% higher than the classical architecture (CNN-LSTM). This model provides a design idea for improving the accuracy of MI-EEG classification and recognition, and also lays a foundation for the wide application of MI-BCI.

Motor Imagery-based Brain-computer interface (MI-BCI) enables communications between the human brain and external machines, which is considered a valuable frontier technology. Motor imagery (MI) identification is a fundamental problem for these applications. However, because of the non-stationary and nonlinear properties of EEG motor imagery signals, it is very hard to get a high accuracy multi-class method. In this study, a multi-class classification method of MI EEG signals using wavelet packets and a one-dimensional convolutional neural network is presented. Data normalization and wavelet threshold denoising were performed for signal preprocessing. Then EEG signals were decomposed and reconstructed by wavelet packets transform. According to the features between different channels of EEG signal, a multilayer one-dimensional convolutional neural network model was adopted to get a multi-class classification model for MI. This model was validated on the public datasets BCI competition IV 2a. The result shows that the mean accuracy of this model is 93.05%. This study is beneficial to MI-BCI research applications in the future.

The brain-computer interface is a novel tool for interaction between the brain and external devices, and is an important tool at the forefront of international brain science research. Motor imagery does not require external stimulation and is one of the most widely used BCI paradigms, and is now widely used to address the treatment of neurological impairments such as paralysis, stroke, and acromegaly. Electrophysiological source imaging is a technique based on the observation of electromagnetic signals from the cerebral cortex, reflecting the electrophysiological activity of brain nerve cells on the scalp surface of the brain. The wavelet transform, on the other hand, is commonly used for the processing of non-smooth signals and aims to extract excellent signal features. This paper combines these two algorithms as feature extraction methods. In recent years, deep learning has been increasingly combined with MI-EEG-based BCI. GoogLeNet is one of the most classical convolutional neural networks and has made a splash in ImageNet image recognition. However, using GoogLeNet to train the BCI public dataset BCI IV IIa and dataset HGD has shown problems such as long training time, overfitting, and gradient disappearance. In response, this paper proposes an improved and streamlined version of GoogLeNet for training based on this dataset to achieve an overall improvement in accuracy, training time, number of parameters, and other indicators. It is useful for the realistic application of MI-based BCI systems.

Brain-Computer Interface (BCI) systems establish a control and communication relationship between the human brain and computers including robots or other devices to help individuals with severe motor disabilities. The classification of motor and mental imagery electroencephalogram (EEG) signals is complicated because these signals are usually case-specific and distinct models must be trained for each subject to process and classify his/her EEG signals. Moreover, in BCI systems EEG signals are processed online, so the time latency must be very low. In this paper, we have proposed a method based on signal-to-image conversion to investigate image processing techniques in the pair-wise classification of motor and mental imagery EEG signals. We first decomposed EEG signals of each trial into four sub-bands. Then, for each sub-band, we converted EEG time series to 2-dimensional (2D) images using covariance between signals of all channels. Then, statistical, textural and PCA-based features were extracted from these images and fed to a support vector machine (SVM) classifier. Our results were promising in the offline processing and achieved an average classification accuracy of 79.57%.