Page 52 - IJEEE-2023-Vol19-ISSUE-1
P. 52

48 |                                                                                     Al-Qaysi1, Al-Saegh, Hussein & Ahmed

minimize the cost of data processing by recognizing the most           It is critical to select the ideal mother wavelet along the
relevant feature components embedded in the signal.                course of using WT with deep transfer learning. The
                                                                   selection of a mother wavelet (MWT) function, on the other
    In this respect, the feature extraction process can be         hand, has been reported in the literature as an important step
conducted in multiple signal processing domains, such as           and component of wavelet analysis to demonstrate the
spatial domain, time domain, frequency domain, and time-           advantages of WT in denoising, component separation,
frequency domain. Particularly, time-frequency features of         coefficient reconstruction, and feature extraction from
MI EEG signals are widely used for classification in BCI           signals in the time and frequency domains [17]. The need for
applications, whereas it describes the density and intensity of    this phase arises from the fact that no unique MWT basis
energy of signal at a different time and frequency by              functions have yet been identified that cater to all EEG
designing a joint function of time and frequency [1].              channels.
Principally, EEG signal analysis in the time-frequency
domain based on wavelet transform (WT) das proved their                Specifically, the research proposes an approach for
ability and usefulness in handling brain signal characteristics    feature extraction and classification that is based on a
compared to other methods such as short-time Fourier               continuous wavelet transform (CWT) in conjunction with
transform (STFT), autoregressive model (ARM), and                  deep learning-based transfer for feature extraction and
wavelet transform (WT) [8]. To date, great attention is given      artificial neural network (ANN) for classification.
to the WT in the field of biomedical signal processing
because of its efficiency in the diagnostic as well as in the          The following is the structure of the reminder for this
pattern recognition [1].                                           paper: WTNN development, evaluation, and validation were
                                                                   all carried out under a methodological framework defined in
    Deep neural networks (DNNs) have recently                      Section 2. Sections 3 and 4, which summarise the findings
demonstrated impressive categorization capabilities in a           and commentary for the experimental component,
variety of applications, including computer vision, video          respectively, are based on two separate datasets, namely the
processing, and speech recognition. Several academics were         BCI Competition dataset IV/2b and the Emotive EPOC
inspired by its enormous success to investigate how deep           dataset, and are divided into two sections. After that, the final
neural networks may be used to categorize EEG signals [9,          section of this study displays the findings of this research.
10]. Researchers have begun using deep learning in their BCI
applications, including seizure detection, memory retrieval,                           II. LITERATURE REVIEW
and MI categorization [11]. The convolution neural network
(CNN) has demonstrated that it is capable of extracting                Al-Qazzaz reported in [18] that even a 0.1%
spatial and temporal characteristics from magnetic induction       intensification in the classification accuracy in medical
(MI) data. It has been shown that CNN can extract excellent        research fields is considered vital due to the high complexity
features using both shallow and deep models, indicating that       of their signals. Therefore, many methods are proposed to
significant features may be retrieved at different levels [4].     attain the highest classification accuracy.

    However, one of the most significant challenges in the             Xu et. al. [19] used a transference CNN framework based
categorization of MI EEG characteristics using deep learning       on VGG-16 and time-frequency spectrum images generated
algorithms is the limited amount of data available due to the      by STFT for analyzing MI EEG signals. They obtained
exhaustion of patients throughout the experiments [7]. There       71.2% of BCI competition classification accuracy for the IV
are also significant individual variations between various         2b dataset. Using a continuous wavelet transform (CWT)
subjects, making it hard to directly utilize the labeled data      filter bank to classify four MI tasks (left hand, right hand,
from other subjects to train the classifier that would be used     feet, and tongue).
to identify the target individuals [12]. Meanwhile, the
collection of EEG data is extremely costly, and a sufficient           Mahamune and Laskar [20] proposed a framework for
quantity of labeled samples is difficult to come by [13].          developing two-dimensional (2D) images for CNNs that uses
When it comes to combining data from domains with                  a continuous wavelet transform (CWT) filter bank to classify
different distributions, transfer learning has emerged as a        four MI tasks. On the BCI competition IV 2a dataset, they
viable approach. Incorporated within transfer learning are         achieved an accuracy of 71.25 percent in classification
methods that are designed to transfer representations and          accuracy. Using a common spatial pattern (CSP) and
information from one domain to another [14]. In other              filtering in conjunction with the empirical mode
words, the approach enables researchers to easily integrate        decomposition (EMD).
fresh datasets into a machine learning model that has already
been trained. Having this functionality can be especially              Alvarez-Meza et al. [21] were able to distinguish the mu
useful in a BCI system since the amount of data supplied is        and beta rhythms from one another. The support vector
frequently insufficient to ensure the appropriate training of a    machine (SVM) classifier was utilized by the researchers.
machine learning model [15]. In the BCI studies, it was            They attained classification accuracy of 92.86 percent on the
discovered that the CNN-based subject-transfer technique           BCIC IV-I dataset and 72.30 percent on the BCIC IV 2b
outperformed the others. Subject-transfer strategies are           dataset, respectively, on the two datasets.
based on the idea that the typical patterns of the target subject
and other subjects may be comparable when doing the same               Wang et. al. [22] proposed a method based on CSP as
activity [16].                                                     preprocessing while feature extraction was done using
                                                                   autoregressive and log-variance; the Kullback-Leibler
                                                                   divergence was for feature selection and time segment
                                                                   selection. Th classification is achieved using the linear
                                                                   discriminate analysis technique. They achieved a
   47   48   49   50   51   52   53   54   55   56   57