One challenge in EEG motor imaging is th e low signal-to-noise ratio of brain signals. Its emergence in the accurate rendition of brain signals varies significantly from person to person. Here, we propose a framework to classify tasks based on fusion features using a Support Vector Machine. Our features are acquired from Discrete Wavelet Transform and Empirical Mode Decomposition. Subsequently, the disparity between measurements of left and right brain signals was calculated. Our proposed work significantly improves accuracy from 83.29 % to 93.16 % compared to previous work.
PL
Jednym z wyzwań w obrazowaniu motorycznym EEG jest niski stosunek sygnału do szumu sygnałów mózgowych. Jego pojawienie się w dokładnym przekazywaniu sygnałów mózgowych różni się znacznie w zależności od osoby. Tutaj proponujemy ramy do klasyfikowania zadań w oparciu o funkcje fuzji przy użyciu maszyny wektorów nośnych. Nasze funkcje są uzyskiwane z dyskretnej transformacji falkowej i dekompozycji trybu empirycznego. Następnie obliczono rozbieżność między pomiarami sygnałów lewego i prawego mózgu. Nasza proponowana praca znacznie poprawia dokładność z 83,29% do 93,16% w porównaniu z poprzednią pracą.
Empirical mode decomposition (EMD) is a popular, user-friendly, data-driven algorithm to decompose a given (non-stationary) signal into its constituting components, utilizing spline interpolation. This algorithm was first proposed in 1998 in the one-dimensional setting, and it employed standard cubic spline interpolation. Since then, different two-dimensional extensions of EMD have been proposed. In this paper, we consider one of these two-dimensional extensions and adapt it to use a shape-preserving interpolation scheme based on quadratic B-splines, ensuring that monotonicity and concavity in the input data are preserved. Using multiple numerical experiments, we show that this new scheme outperforms the original EMD, both qualitatively and quantitatively.
The empirical mode decomposition (EMD) algorithm is widely used as an adaptive time-frequency analysis method to decompose nonlinear and non-stationary signals into sets of intrinsic mode functions (IMFs). In the traditional EMD, the lower and upper envelopes should interpolate the minimum and maximum points of the signal, respectively. In this paper, an improved EMD method is proposed based on the new interpolation points, which are special inflection points (SIPn) of the signal. These points are identified in the signal and its first (n − 1) derivatives and are considered as auxiliary interpolation points in addition to the extrema. Therefore, the upper and lower envelopes should not only pass through the extrema but also these SIPn sets of points. By adding each set of SIPi (i = 1, 2, n) to the interpolation points, the frequency resolution of EMD is improved to a certain extent. The effectiveness of the proposed SIPn-EMD is validated by the decomposition of synthetic and experimental bearing vibration signals.
To ensure that any time series data is appropriately interpreted, it should be analyzed with proper signal processing tools. The most common analysis methods are kernel-based transforms, which use base functions and their modifications to represent time series data. This work discusses an analysis of audio data and two of those transforms - the Fourier transform and the wavelet transform based on a priori assumptions about the signal's linearity and stationarity. In audio engineering, these assumptions are invalid because the statistical parameters of most audio signals change with time and cannot be treated as an output of the LTI system. That is why recent approaches involve decomposition of a signal into different modes in a data-dependent and adaptive way, which may provide advantages over kernel-based transforms. Examples of such methods include empirical mode decomposition (EMD), ensemble EMD (EEMD), variational mode decomposition (VMD), or singular spectrum analysis (SSA). Simulations were performed with speech signal for kernel-based and data-dependent decomposition methods, which revealed that evaluated decomposition methods are promising approaches to analyzing non-stationary audio data.
This paper introduces the notion of a generalized measure for a sequence of functions with oscillation and concentration effects. This measure is constructed by averaging the sequence of Borel measurable functions using singular or regular perturbations. In this way, the generalized limits of such sequences are conceptualized by enlarging the space of functions to measure spaces. It is a modification of the Young measure. This modified measure was termed a Q-measure. It can be difficult to determine the Young measure for a broad function. The Q-measure can be easily calculated for particular functions. This is one of the advantages of this study. As an application of the measure, we can define another weaker type of Monotone convergence theorem, the Lebesgue-dominated convergent theorem. A notion of average for underlying sequences to define the Q-measure is given, as also its application in signal analysis and atmospheric sciences.
PL
W tym artykule autor wprowadza nową miarę, którą nazywa miarą Q, reprezentującą słabą∗ granicę barycentrum ciągu funkcji borelowskich. Omawia niektóre wyniki zwi¡zane z tą miarą, co jest pomocne przy wyznaczaniu miary Q dla poszczególnych typów funkcji. Ponadto omówiono zastosowanie koncepcji średniej w analizie sygnałów i naukach o atmosferze.
The Q-measure indicate a weak∗ limit of the barycenter of a sequence of Borel measurable functions. In this paper, we will look only at Q-functional. Q-functional is defined by Q- measure, it is useful in the field of optimization. Computational results for Q-functional are presented and compared with Young functional. The obtained analytical results demonstrate relative error in Q-functional is lesser compared to Young functional.
PL
Miara Q wyznacza słabą∗ granicę barycentrum ciągu funkcji borelowskich. W tym artykule przyjrzymy się tylko funkcjonałom Q. Q-funkcjonalność jest definiowana przez miarę Q i jest przydatna w zastosowaniu do zadań optymalizacji. Przedstawiono wyniki obliczeń dla funkcjonału Q i porównano je z funkcjonałem Younga. Otrzymane wyniki analityczne pokazują, że błąd względny w funkcjonale Q jest mniejszy w porównaniu z funkcjonałem Younga.
7
Dostęp do pełnego tekstu na zewnętrznej witrynie WWW
A reliable computer-aided method for Parkinson’s disease (PD) detection can slow down its progression and improve the life quality of patients. In this study, a new non-invasive and cost-effective method based on the online analysis of handwriting signals has been proposed. First, the dynamic handwriting signals have been converted into two graphical representations of the variability rate. Then, two new feasible features, including area of the analytic signal representation and area of the second-order difference plot, have been used to quantify the variability rate of handwriting signals. A statistical test and support vector machine classifier have been applied in a comparative study to test the impact of each variability feature, writing task, and time sequence on the detection performance, separately. The obtained results on PaHaW database with 35 Parkinson’s disease patients and 36 healthy controls have shown that the proposed method of handwriting variability feature extraction has effective performance and the capability for the PD detection. It has achieved an average sensitivity of 86.26% with only two types of features, providing a trade-off between the performance, the computational complexity, and interpretability of the motor patterns from the point of view of clinicians and neuropsychologists. Xcoordinate time-series and writing a sentence can achieve superior accuracy and robustness in the presence of individual differences. The experimental results have demonstrated that extracting the variability features that used graphical representations of the global changes in oscillatory mode has the ability to clinically describe the pathological dynamics of the handwriting signals for the PD identification.
8
Dostęp do pełnego tekstu na zewnętrznej witrynie WWW
The excessive drinking of alcohol can disrupt the neural system. This can be observed by properly analysing the Electroencephalogram (EEG) signals. However, the EEG is a signal of complex nature. Therefore, an accurate categorization between alcoholic (A) and nonalcoholic (NA) subjects, while using a short time EEG recording, is a challenging task. In this paper a novel hybridization of the oscillatory modes decomposition, features mining based on the Second Order Difference Plots (SODPs) of oscillatory modes, and machine learning algorithms is devised for an effective identification of alcoholism. The Empirical Mode Decomposition (EMD) and Variational Mode Decomposition (VMD) are used to respectively decompose the considered EEG signals in Intrinsic Mode Functions (IMFs) and Modes. Onward, the SODPs, derived from first six IMFs and Modes, are considered. Features of SODPs are mined. To reduce the dimension of features set and computational complexity of the classification model, the pertinent features selection is made on the basis of Wilcoxon statistical test. Three features with p-values (p) of < 0.05 are selected from each intended SODP and these are the Central Tendency Measure (CTM), area and mean. These features are used for the discrimination between A and NA classes. In order to determine a suitable EEG signal segment length for the intended application, experiments are performed by considering features extracted from three different length time windows. The classification is carried out by using the Least Square Support Vector Machine (LS-SVM), Multilayer perceptron neural network (MLPNN), K-Nearest Neighbour (KNN) and Random Forest (RF) algorithms. The applicability is tested by using the UCI-KDD EEG dataset. The results are noteworthy for MLPNN with 99.89% and 99.45% accuracies for EMD and VMD respectively for 8-second window.
Time series models have been used to extract damage features in the measured structural response. In order to better extract the sensitive features in the signal and detect structural damage, this paper proposes a damage identification method that combines empirical mode decomposition (EMD) and Autoregressive Integrated Moving Average (ARIMA) models. EMD decomposes nonlinear and non-stationary signals into different intrinsic mode functions (IMFs) according to frequency. IMF reduces the complexity of the signal and makes it easier to extract damage-sensitive features (DSF). The ARIMA model is used to extract damage sensitive features in IMF signals. The damage sensitive characteristic value of each node is used to analyze the location and damage degree of the damaged structure of the bridge. Considering that there are usually multiple failures in the actual engineering structure, this paper focuses on analysing the location and damage degree of multi-damaged bridge structures. A 6-meter-long multi-destructive steel-whole vibration experiment proved the state of the method. Meanwhile, the other two damage identification methods are compared. The results demonstrate that the DSF can effectively identify the damage location of the structure, and the accuracy rate has increased by 22.98% and 18.4% on average respectively.
Heart rate is constantly changing under the influence of many control signals, as manifested by heart rate variability (HRV). HRV is a nonstationary, irregularly sampled signal, the spectrum of which reveals distinct bands of high, low, very low and ultra-low frequencies (HF, LF, VLF, ULF). VLF and ULF components are the least understood, and their analysis requires HRV records lasting many hours. Moreover, there are still no well-established methods for the reliable extraction of these components. The aim of this work was to select, implement and compare methods which can solve this problem. The performance of multiband filtering (MBF), empirical mode decomposition and the short-time Fourier transform was tested, using synthetic HRV as the ground truth for methods evaluation as well as real data of three patients selected from 25 polysomnographic records with a clear HF component in their spectrograms. The study provided new insights into the components of long-term HRV, including the character of its amplitude and frequency modulation obtained with the Hilbert transform. In addition, the reliability of the extracted HF, LF, VLF and ULF waveforms was demonstrated, and MBF turned out to be the most accurate method, though the signal is strongly nonstationary. The possibility of isolating such waveforms is of great importance both in physiology and pathophysiology, as well as in the automation of medical diagnostics based on HRV.
Tool wear condition monitoring (TCM) is essential for milling process to ensure the machining quality, and the long short-term memory network (LSTM) is a good choice for predicting tool wear value. However, the robustness of LSTM- based method is poor when cutting condition changes. A novel method based on data fusion enhanced LSTM is proposed to estimate tool wear value under different cutting conditions. Firstly, vibration time series signal collected from milling process are transformed to feature space through empirical mode decomposition, variational mode decomposition and fourier synchro squeezed transform. And then few feature series are selected by neighborhood component analysis to reduce dimension of the signal features. Finally, these selected feature series are input to train the bidirectional LSTM network and estimate tool wear value. Applications of the proposed method to milling TCM experiments demonstrate it outperforms significantly SVR- based and RNN- based methods under different cutting conditions.
12
Dostęp do pełnego tekstu na zewnętrznej witrynie WWW
Voice disorders are one of the incipient symptoms of Parkinson's disease (PD). Recent studies have shown that approximately 90% of PD patients suffer from vocal disorders. Therefore, it is significant to extract pathological information on the voice signals to detect PD. In this paper, a feature, named energy direction features based on empirical mode decomposition (EDF-EMD), is proposed to show the different characteristics of voice signals between PD patients and healthy subjects. Firstly, the intrinsic mode functions (IMFs) were obtained through the decomposition of voice signals by EMD. Then, the EDF is obtained by calculating the directional derivatives of the energy spectrum of each IMFs. Finally, the performance of the proposed feature is verified on two different datasets: dataset-Sakar and dataset-CPPDD. The proposed approach shows the best average resulting accuracy of 96.54% on dataset-Sakar and 92.59% on dataset-CPPDD. The results demonstrate that the method proposed in this paper is promising in the field of PD detection.
13
Dostęp do pełnego tekstu na zewnętrznej witrynie WWW
Alcoholism can be analyzed by Electroencephalogram (EEG) data. Finding an optimal subset of EEG channels for alcoholism detection is a challenging task. The paper reports a new methodology for the detection of optimal channels for alcoholism analysis using EEG data. The proposed technique employs the Empirical Mode Decomposition (EMD) technique to extract the amplitude and frequency modulated bandwidth features from the Intrinsic Mode Function (IMF) and ensemble subspace K-NN as a classifier to classify alcoholics and normal. The optimum channels are selected, using a harmony search algorithm. The fitness value of discrete binary harmony search (DBHS) optimization algorithms is calculated using accuracy and sensitivity achieved by the ensemble subspace K-Nearest Neighbor classifier. Experimental outcomes indicate that the optimal channel selected by the harmony search algorithm has biological inference related to the alcoholic subject. The proposed approach reports a classification accuracy of 93.87%, with only 12 detected EEG channels.
14
Dostęp do pełnego tekstu na zewnętrznej witrynie WWW
Epilepsy is a brain disorder in which patients undergo frequent seizures. Around 30% of patients affected with epilepsy cannot be treated with medicines/surgical procedures. Abnormal activity, known as the preictal state starts few minutes before the seizure actually occurs. Therefore, it may be possible to deliver medication prior to the occurrence of a seizure if initiation of the preictal state can predicted before the seizure onset. We propose an epileptic seizure prediction method that predicts the preictal state before the seizure onset using electroencephalogram (EEG) monitoring of brain activity. It involves three steps including preprocessing of EEG signals, feature extraction classification of preictal and interictal states. In our proposed method, we have used (i) Empirical model decomposition to remove noise from the EEG signals and Generative Adversarial Networks to generate preictal samples to deal with the class imbalance problem; (ii) Automated features have been extracted with three layer Convolutional Neural Networks and (iii) Classification between preictal and interictal states is done with Long Short Term Memory units. In this study, we have used CHBMIT dataset of scalp EEG signals and have validated our proposed method on 22 subjects of dataset. Our proposed seizure prediction method is able to achieve 93% sensitivity and 92.5% specificity with average time of 32 min to predict the seizure's onset. Results obtained from our method have been compared with recent state-of-the-art epileptic seizure prediction methods. Our proposed method performs better in terms of sensitivity, specificity and average anticipation time.
15
Dostęp do pełnego tekstu na zewnętrznej witrynie WWW
The separation of electrooculogram (EOG) and electroencephalogram (EEG) is a potential problem in brain-computer interface (BCI). Especially, it is necessary to accurately remove EOG, as a disturbance, from the measured EEG in brain disease diagnosis, EEG-based rehabilitation systems, etc. Due to the interaction between the eye and periocular musculature, a multipoint spike is often produced in EEG for each ocular activity. Masking-aided minimum arclength empirical mode decomposition (MAMA-EMD) was developed to robustly decompose time series with impulse-like noise. However, the decomposition performance of MAMA-EMD was limited in the case of one impulse with multiple contiguous spike points. In this paper, MAMA-EMD was improved (called IMAMA-EMD) by supplementing the minimum arclength criterion, and it was combined with kernel independent component analysis (KICA), yielding an automatic EOG artifact removal method, denoted as KIIMME. The multi-channel contaminated EEG signals were separated into several independent components (ICs) by KICA. Then, IMAMA-EMD was applied to the EOG-related ICs decomposition to generate a set of inherent mode functions (IMFs), the low frequency ones, which have higher correlation with EOG components, were removed, and the others were employed to construct ‘clean’ EEG. The proposed KIIMME was evaluated and compared with other methods on semisimulated and real EEG data. Experimental results demonstrated that IMAMA-EMD effectively eliminated the influence of multipoint spike on sifting process, and KIIMME improved the removal accuracy of EOG artifacts from EEG while retaining more useful neural data. This improvement is of great significance to research on brain science as well as BCI.
16
Dostęp do pełnego tekstu na zewnętrznej witrynie WWW
Parkinson's disease (PD) is a progressive neurological disorder prevalent in old age. Past studies have shown that speech can be used as an early marker for identification of PD. It affects a number of speech components such as phonation, speech intensity, articulation, and respiration, which alters the speech intelligibility. Speech feature extraction and classification always have been challenging tasks due to the existence of non-stationary and discontinuity in the speech signal. In this study, empirical mode decomposition (EMD) based features are demonstrated to capture the speech characteristics. A new feature, intrinsic mode function cepstral coefficient (IMFCC) is proposed to efficiently represent the characteristics of Parkinson speech. The performances of proposed features are assessed with two different datasets: dataset-1 and dataset-2 each having 20 normal and 25 Parkinson affected peoples. From the results, it is demonstrated that the proposed intrinsic mode function cepstral coefficient feature provides superior classification accuracy in both data-sets. There is a significant increase of 10–20% in accuracy compared to the standard acoustic and Mel-frequency cepstral coefficient (MFCC) features.
17
Dostęp do pełnego tekstu na zewnętrznej witrynie WWW
The recognition of human activities is a topic of great relevance due to its wide range of applications. Different approaches have been proposed to recognize human activities, ranging from the comparison of signals with thresholds to the application of deep and machine learning techniques. In this work, the classification of six human activities (walking, walking downstairs, walking upstairs, standing, sitting, and lying down) is performed using bidirectional LSTM networks that exploit intrinsic mode function (IMF) representation of inertial signals. Records with inertial signals (accelerometer and gyroscope) of 2.56 s, available at the UCI Machine Learning Repository, were collected from 30 subjects using a smartphone. First, inertial signals were standardized to take them to the same scale and were decomposed into IMF using the improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN). IMF were then segmented (split) into nine segments of 1.28 s with 12.5% overlap and introduced to a first network with four outputs to identify the dynamic activities and the statics as a single class called ‘‘statics’’, giving 98.86% accuracy. Then, the non-segmented IMF of the records assigned to the statics class were introduced to a second network to classify their three activities, giving an accuracy of 88.46%. In total, 92.91% accuracy was obtained to classify the six human activities. This performance is because ICEEMDAN allowed the extraction of information that was embedded in the signal, and the segmentation of the IMF allowed the network to discriminate between static and dynamic activities.
18
Dostęp do pełnego tekstu na zewnętrznej witrynie WWW
Phonocardiogram (PCG) recordings contain valuable information about the functioning and state of the heart that is useful in the diagnosis of cardiovascular diseases. The first heart sound (S1) and the second heart sound (S2), produced by the closing of the atrioventricular valves and the closing of the semilunar valves, respectively, are the fundamental sounds of the heart. The similarity in morphology and duration of these heart sounds and their superposition in the frequency domain makes it difficult to use them in computer systems to provide an automatic diagnosis. Therefore, in this paper, we analyzed these heart sounds in the intrinsic mode functions (IMF) domain, which were issued from two time-frequency decomposition techniques, the empirical mode decomposition (EMD) and the improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN), with the aim of retrieving useful information on an expanded basis. The decomposition of PCG recordings into IMF allows representing the fundamental cardiac sounds in many oscillating components, increasing thus the observability of the system. Moreover, the time-frequency representation of PCG recordings could provide valuable information to automatically detect heart sounds and diagnose pathologies from characteristic patterns of these heart sounds in the IMF. The analysis was made through the variance and Shannon's entropy of the heart sounds, observed in time windows located among different IMF. In addition, we determined the frequencies ranges of the IMF from the decomposition of the PCG recordings using both techniques. Given that the frequency content of S1 and S2 is different but overlap each other, and the duration of these sounds are also different, these heart sounds were represented in different IMF with different variances and entropies, in both techniques, but the ICEEMDAN offers a more consistent decomposition of S1 and S2 (they were concentrated in IMF 4-6). The decomposition of PCG signals into IMF has allowed us to identify the frequency components of the IMF in which these sounds are found.
19
Dostęp do pełnego tekstu na zewnętrznej witrynie WWW
Steady-state visual evoked potential (SSVEP) based brain–computer interfaces have been widely studied because these systems have potential to restore capabilities of communication and control of disable people. Identifying target frequency using SSVEP signals is still a great challenge due to the poor signal-to-noise ratio of these signals. Commonly, this task is carried out with detection algorithms such as bank of frequency-selective filters and canonical correlation analysis. This work proposes a novel method for the detection of SSVEP that combines the empirical mode decomposition (EMD) and a power spectral peak analysis (PSPA). The proposed EMD+PSPA method was evaluated with two EEG datasets, and was compared with the widely used FB and CCA. The first dataset is freely available and consists of three flickering light sources; the second dataset was constructed and consists of six flickering light sources. The results showed that proposed method was able to detect SSVEP with high accuracy (93.67 ± 9.97 and 78.19 ± 23.20 for the two datasets). Furthermore, the detection accuracy results achieved with the first dataset showed that EMD+PSPA provided the highest detection accuracy (DA) in the largest number of participants (three out of five), and that the average DA across all participant was 93.67 ± 9.97 which is 7% and 4% more than the average DA achieved with FB and CCA, respectively.
The modern speech processing techniques enable new possibilities of potential applications. Besides speech and speaker recognition, also the information about speakers’ physical condition, emotional state or stress can be detected in speech signal. Since emotional stress can occur during deception, its detection in speech could be used for law or security services. The paper presents the comparative tests of two voice stress detection techniques: one based on trials of microtremors detection relying on an iterative EMD method (Empirical Mode Decomposition) and the second one based on the statistical analysis of fundamental frequency and MFCC parameters. The preliminary tests were carried on the group of 12 speakers (6 males and 6 females) answering yes/no to the list of a few dozen personal questions. The presented research revealed the speakers’ very high personal influence on the obtained results.
JavaScript jest wyłączony w Twojej przeglądarce internetowej. Włącz go, a następnie odśwież stronę, aby móc w pełni z niej korzystać.