34; Frequency recognition of selected features of EEG signals by neuro-statistical method" has been approved by the board of examiners for partial fulfillment of the requirements for the degree of Master of Science in the Department of Electrical and Electronic Engineering, Khulna University of Engineering & Performance of ICA depends on the nature of the task and the algorithm used.
Feature Selection (FS)
NCCA is then applied to the EEG data to search for salient features, where statistical feature set analysis is used without assuming the default learning model as a filter approach [ 36 ].
Frequency Recognition
These selected features are presented in three different CCA networks one by one to recognize frequencies. Comparative analysis of recognition rates from selected and non-selected features is also analyzed with three different CCA networks at three harmonic situations.
Organization of the thesis
Both the optimized signals carry the information of subject-specific and trial-to-trial variability, because these are created from EEG signals. It measures voltage fluctuations due to ionic current flow within the neurons of the brain [38].
Electroencephalography (EEG)
Most of the cerebral signal observed in scalp EEG falls in the range of 1–20 Hz (activity below or above this range is likely to be artifactual under standard clinical recording techniques). The change in the ongoing EEG due to these stimuli is called event-related potential (ERP), in the case of external stimulation also called evoked potential (EP).
Data Collection
Experimental Orientations
Steady-State Visual Evoked Potentials (SSVEP) are brain responses that are precisely synchronized with rapid repetitive external visual stimulation such as flashes, reversal patterns, or luminance-modulated images. When the retina is excited by a visual stimulus ranging from 3.5 Hz to 75 Hz [52], the brain generates electrical activity at the same (or multiples of) frequency of the visual stimulus. These responses can be measured within narrow frequency bands (such as ±0.1 Hz) around the visual stimulation frequency or using other signal processing methods that take advantage of the specific characteristics of the SSVEP signal, such as rhythmicity and synchronization.
While some of the variation in reported frequency dominance may be due to the diversity of brain processes and regions captured depending on the recording modality, SSVEP stimulus parameters such as spatial frequency, luminance, contrast, and color also play a role. decisive. Regan showed that checkerboard stimuli with patterns with small controls (such as 0.2° arc side) exhibit low-frequency preferences with response peaks at −7 Hz, whereas patterns with larger controls (such as arc side 0.7°) have a higher frequency preference, similarly to unpatterned vibration stimuli [53]. A chin rest was used by all subjects to prevent excessive contamination of the EEG data with EMG artifacts due to upper body muscle movements.
Characteristics of data
Page 116 In this thesis, frequency components of high-dimensional SSVEP data are extracted with low computational cost using neuro-statistical method. Visual stimulations are used to flicker the checkerboard and at the same time EEG signals are recorded from the brain with 128 active electrodes.
Aspects of Frequency Recognition
Distances between points in appropriate embeddings of the data are used to compute a set of metric properties. Mayer-Kress and Layne [57] used the EEG time series reconstruction techniques to obtain their phase portrait. The SSVEP is a periodic response to a visual stimulus which has the same ftindamental frequency as that of the visual stimulus as well as its harmonics.
SSVEP can be recorded from the surface of the scalp above the visual cortex. A method that can recognize the frequency with a harmonic ratio can greatly improve the performance of SSVEP recognition. A series of function groups are used instead of the traditional use of individual functions. ii).
Neuro-Statistical Technique
The dimensionality of these new bases is equal to or less than the smallest dimensionality of the two variables. This problem can be considered to maximize a function as g () = E(y1 y2), which is defined as a function of weights, Wj with respect to another set of parameters, w2. The relative strength of the constraint compared to the optimization function is changed by varying the Lagrange multipliers in proportion to the derivatives of J .
The dimensionality of these new bases is equal to or less than the smaller dimensionality of the two variables. CCA can also be defined as the problem of finding two sets of basis vectors, one for x and one for y, such that the correlations between the projections of the variables onto these basis vectors are mutually maximized. Only one of the Eigen value equations needs to be solved since the solutions are related to .
The Framework
The total covariance matrix is a block matrix, where C and ci,), are the covariance matrices x and y within sets, respectively, and CXY =Cyx is the covariance matrix between sets. The number of non-zero solutions of these equations is limited to the smallest dimension of x and y. For example, if the dimensionality of x and y is 8 and 5, respectively, the maximum number of canonical correlations is 5.
L NCCA
Feature Selection Procedure
- Clustering using wavelet
- Data Preparation
- Reference Signal
- Implementation of Colin's CCA Network
First, the entire data set is subdivided into three subsections and they are fed into the CCA network sample by sample. The whole methodology for frequency recognition is investigated in three subsections respectively as reference signal, Cohn's CCA network and extraction of frequency components. Since pure sine-cosine reference signals do not contain any information about EEG data, we generate this kind of optimized reference signals with CCA network.
The entire row-by-row column of specified objects/reference signals is fed into the CCA network at time as input (xi, x2). Here we visualize how the frequency components of EEG signals are extracted using a CCA network. In this case, we implemented a two-stage CCA network that favors the full optimization of the EEG signal with the reference signal.
Experimental Setup
Initially the evaluations of the selected features are done using NN where the data is divided into three subsections before FS. In this case, the network gives the output corresponding to a learned input pattern that is least different from the given pattern. There are 7, 15, and 30 hidden neurons in a hidden layer for 15, 30, and 60 samples, respectively, and they send data via synapses to the output layer of the output neurons.
It appears from Table 5.1 that for subject 4 at 8 Hz, the classifier cannot classify nine attributes. It appears from Table 4.2 that there is no coincidence when testing at 14 Hz stimulation, but trained at 8 Hz stimulation; because all test data are misclassified with training set. So we can say that although brain signals are time-varying quantities, it is possible to test the coincidence of signals from selected functions in the CCA network.
Results of Frequency Recognition (FR)
- FR from Original EEG data
Page 143 Figure 4.5: Correlation coefficient of NCCA with (a) linear network, (b) nonlinear network and. c) nonlinear network with feedback for Hi settings.
- Subjects Variability Realization
- FR from Selected Features
It can be seen that almost the same correlation for each trial of a subject with sine-cosine signals as shown in figure. Three harmonic conditions of sine-cosine reference signals were analyzed for each selected set of features. The highest correlation is found at 1 Hz for each selected feature of S3 in Hi condition, as shown in Figure 2.
In the H2 condition, the dictated frequency is found to be 1 Hz for each subject as shown in Fig. The maximum correlation was obtained at 1 Hz with the Hi condition for each subject as an example seen from Fig. Although the effects of harmonics are different for different subjects, it has also been found that the maximum correlation occurs at the stimulus frequency I Hz for each subject.
L1LLLI1L
Comparative Study
- Correlation Point of View
In this sense, the network is suitable for finding known frequency and can differentiate between subjects by observing correlation profiles. Hz for all the above cases, but negligible correlations are found at other frequencies for the selected features, but not for the original EEG data. In this sense, the expected frequency can be easily obtained from the correlation profiles of the selected EEG features rather than the original one.
L!LLLHL [L-W-
LLLLLL
Computational Complexity
Therefore, the CCA method often does not result in the optimal recognition accuracy of the SSVEP frequency due to possible overfitting. The multi-way CCA method [12] (using tensor data) has shown unproven SSVEP frequency recognition performance compared to the CCA method. On the other hand, a phase-constrained CCA (PCCA) method [106] has also been proposed for SSVEP frequency recognition.
However, the reference signal optimization procedures in both the multidirectional CCA and PCCA methods are not fully based on the training data, but still need to address the pre-constructed sine-cosine waves. Although the multi-group CCA method is fully based on the training data and performs better than various CCA methods for SSVEP recognition, but it is only suitable for a small number of channels. This method can clearly solve the above problems because it is completely based on the training data and reduces the computational time and cost.
Conclusions
Future Works
Du A GA-Based Approach to ICA Feature Selection: An Efficient Method to Classify Microarray Datasets". Destexhe Electroencephalogram." in Proceedings of the IEEE First International Conference on Neural Networks, edited by M.
Pal A neuro-fuzzy scheme for simultaneous feature selection and fuzzy rule-based classification "IEEE Transactions on Neural Networks, vol. Ding Feature selection based on mutual information: criteria of maximum dependence, maximum relevance and mm-redundancy" . Rui Sparse Transfer Learning for Interactive Video Query Reordering,” in: Proceedings of the ACM Transactions on Multimedia Computing, Communications and Applications.
