Earticle

In this study, contribution evaluation method applying Independent Component Analysis (ICA) was proposed. The necessity of applying ICA to the contribution evaluation was investigated through numerical simulation. The simulation modeled a scenario where the vibration/noise sources were physically overlapped in a small space, and their frequency characteristics were similar. For comparison between the conventional contribution evaluation method and the proposed method, the contribution evaluation was performed using the ordinary and partial contribution evaluation methods. Through this analysis, it was confirmed that the proposed method can identify contributions by restoring the signal when the frequency characteristics of the vibration/noise sources were similar, and their positions overlapped. These results confirm that the contribution evaluation method based on independent component analysis is effective in appropriately analyzing vibration/noise sources when their frequency characteristics are similar, and their positions overlap.

독립성분분석(ICA)은 미지의 신호원으로부터 측정된 신호를 분리하기 위한 비감독화 선형 변환법으로 알려져있다. 본 논문의 목표는 ICA기법에 기초한 응답기반 시스템 식별과 현장 구조물에 적용 가능성을 확인하는 것이다. 구조물에 태풍이 접근할 때 구조물의 상부에 설치된 3개의 가속도계로부터 가속도 응답을 직접적으로 측정된다. 측정된 가속도 응답은 구조물의 모드와 밀접히 관련되어 병진-비틀림이 연계되어 나타난다. 모드분리를 위한 성능지수로서 첨도를 최소화하기위해 ICA기법을 적용하여, 병진-비틀림이 연계된 모드 응답은 밀접하게 인접한 모드일지라도 정확하게 분리되고 변환행렬에 의해 자동적으로 모드 형상은 결정되어지는 것을 확인하였다. 본 연구 결과로부터 고유진동수와 같은 모드 특성은 힐버트 변환에서 얻은 각 모드의 시변 고유진동수의 누적 분포 함수로부터 추계학적으로 추정할 수 있다. 본 논문의 결과는 새롭게 제안된 기법이 기존의 응답기반 시스템 식별보다 구조물의 응답에 매우 강력하게 적용되는 것을 보여준다.

Independent component analysis(ICA) is known as an unsupervised linear transformation method usefully applied to separate measured signals into blind source signals. This study aims to propose a new output-only system identification based on the ICA technique and to verify the applicability of the method to in-situ building structure. The three acceleration responses of a structure are directly measured from accelerometers installed at the top of building structure when the typhoon is approaching to the structure. The measured acceleration responses are strongly coupled with the effect of the lateral-torsional motion induced by the closely located modes of the structure. By applying the ICA technique to minimize the kurtosis introduced as the performance index for the mode separation, it is found that the lateral-torsionally coupled modal response can be separated precisely by ICA technique even the modes are closely located, and automatically the mode shape can be determined in the process of the separation. From the results, it is also shown that the modal properties such as natural frequencies can be stochastically estimated from the cumulative distribution function of instantaneous natural frequency of each mode obtained from Hilbert transform. The results in this study show that the new proposed method can be very usefully applied to structural responses than the conventional output-only system identification.

Traditionally, the key idea of estimating independent component analysis (ICA) model is to maximize the non-Gaussianity, however, often with the assumption that density of data is near the standardized Gaussian density. To avoid the unsuitable assumption, this article uses the nonparametric density estimating method. A nonparametric independent component analysis algorithm based on Epanechnikov kernel function is proposed in this paper. This algorithm uses the Epanechnikov kernel estimator to estimate random variable distribution, meanwhile, employs the hypothesis test to derive the nonparametric likelihood ratio (NLR) objective function. For optimizing the nonparametric density estimation, the selection of kernel function and bandwidth is crucial. From the perspective of minimizing the mean integrated square error (MISE), this paper discusses the optimal selection and conducts experiments for further study. To increase the algorithmic convergence rate and reduce the running time, the quasi-newton method has been used to optimize the objective function. Compared with previous nonparametric ICA algorithm, the simulation results demonstrate that the proposed method offers better performance both on speech separation and computing capability.

In ultrasonic nondestructive testing, the presence of noise makes great trouble for defect recognition, so it is very necessary to reduce noise in collected ultrasonic signal. In this paper a de-noising algorithm of ultrasonic echo signal based on wavelet transform and Independent Component Analysis (ICA) was presented. First, wavelet transform was used to decompose original noisy signal, and then ICA was applied to decomposed detail coefficients, separated independent components were evaluated by threshold, noise was filtered, and finally, de-noised ultrasonic signal was obtained by wavelet reconstruction. Simulation and experimental results showed that the proposed algorithm can improve signal-to-noise ratio, meanwhile, overcome some other de-noising algorithms’ shortcoming of losing useful information in de-noising, the performance is superior to wavelet threshold de-noising algorithm.

This paper aims to construct a novel wireless communication system, in which source signals are transmitted simultaneously in the same frequency band. The transmitted signals are only required to be statistically independent or statistically distinguished. Therefore, the source signals can be recovered at the receiver by utilizing the classical algorithms of blind source separation (BSS) and independent component analysis (ICA) such as the fast fixed-point algorithm (FastICA). On the one hand, because the source signals are transmitted simultaneously in the same frequency band, the spectrum efficiency of this novel system is much higher than those of time division multiplexing (TDM), frequency division multiplexing (FDM), and code division multiplexing (CDM) systems, in which TDM, FDM and CDM signals are limited in time interval, frequency band and code. On the other hand, inspired by recently proposed reference-based schemes, the reference signals are introduced to the classical separation algorithms of BSS and ICA, which makes this novel system much more efficient than classical ones in terms of computational speed. The performance of this new system is validated through realistic experiments. Additionally, it is theoretically shown that the information content of all the source signal inputs can be recovered by this novel wireless communication system.

Safety driving performance estimation and alertness (SDPEA) has drawn the attention of researchers in preventing traffic accidents caused by drowsiness while driving. Psychophysiological measures, such as electroencephalogram (EEG), are accurately investigated to be robust candidates for drivers’ drowsiness evaluation. This paper presents an effective EEG-based driver drowsiness monitoring system by analyzing the changes of brain activities in a simulator driving environment. The proposed SDPEA system can translate EEG signals into drowsiness level. Firstly, Independent component analysis (ICA) is performed on EEG data to remove artifacts. Then, eight EEG-band powers- related features: beta, alpha, theta, delta, (alpha plus theta)/beta, alpha / beta, (alpha plus theta)/(alpha plus beta) and theta / beta are extracted from the preprocessed EEG signals by employing the Fast Fourier Transform (FFT). Subsequently, fisher score technique selects the most descriptive features for further classification. Finally, Support Vector Machine (SVM) is employed as a classifier to distinguish drowsiness level. Experimental results show that the quantitative driving performance can be correctly estimated through analyzing driver’s EEG signals by the SDPEA system.

This paper addresses the problem of independent component extraction of complex-valued signals in convolutive mixtures. Most previous research focused on real-valued convolutive ICA, and corresponding solution methods are generally computationally complex and inefficient for real application. In order to solve the problem, we propose a novel method based on first-order statistics, which includes several single-step and iterative separators to satisfy different demands of engineering applications. We also provided the theoretical performance analysis, which is validated by experimental simulations. It is observed from the simulations that various factors (especially the noncircularity) affect the extraction performance of separators; hence we offer some advice on how to choose separators properly. Besides, the proposed iterative separators generally perform better and converge faster compared to two complex FastICA algorithms.

In this paper, a new method based on Particle Swarm Optimization (PSO) for independent component analysis(ICA) is presented which can be applied for feature extraction. Due to the drawbacks of the Gradient method, it is replaced by PSO in Discriminant Independent Component Analysis (dICA ) algorithm in the proposed approach. The Gradient method may lead to local optimal and it cannot solve the problem of slow convergence since it includes a learning step which needs to be determined in advance. Moreover, Gradient-based techniques cannot achieve high level of accuracy because of the considerable complexity involved in ICA. The additional complexity of the Gradient-based algorithms leads to pseudo-optimal scenarios. The Discriminant Independent Component Analysis based on PSO is used to overcome these serious shortcomings. Most of the datasets used for simulation in this study are obtained from UCI repository. The results obtained using linear discriminant analysis (LDA), principal component analysis (PCA) and gradient-based dICA are compared with those obtained by PSO-dICA . The results show improvement in classification with PSO-dICA method compared to other methods. In other words, PSO-dICA method bought about classifier error reduction.

Multispectral approach to brain MRI analysis has shown great advance recently in pathology and tissue analysis. However, poor performance of the feature extraction and classification techniques involved in it discourages radiologists to use it in clinical applications. Transform based feature extraction methods like Independent Component Analysis (ICA) and its variants have contributed a lot in this research field. But these global transforms often fails in extraction of local features like small lesions from clinical cases and noisy data. Feature extraction part of the recently introduced Multiresolution Independent Component Analysis (MICA) algorithm in microarray classification is proposed in this work to resolve this issue. Effectiveness of the algorithm in MRI analysis is demonstrated by training and classification with Support Vector Machines (SVM). Both synthetic and real abnormal data from T1-weighted, T2-weighted, proton density, fluid-attenuated inversion recovery and diffusion weighted MRI sequences are considered for detailed evaluation of the method. Tanimoto index, sensitivity, specificity and accuracy of the classified results are measured and analyzed for brain abnormalities, affected white matter and gray matter tissues in all cases including noisy environment. A detailed comparative study of classification using MICA and ICA is also carried out to confirm the positive effect of the proposed method. MICA based SVM is found to yield very good results in anomaly detection, around 2.5 times improvement in classification accuracy is observed for abnormal data analysis.

fMRI의 신호는 매우 다양한 종류의 선호들이 혼합된 상태이며 , 비록 몇 가지의 요소에 대해 모델링하여 그 선호 형태를 추측할 수 있으나 모든 신호를 정확하게 분리하여 뇌신경의 활성화를 반영하는 신호만을 선택적으로 알아 내기는 어려운 일이다. 또한 뇌와 신체의 생리적 현상으로 발생하는 잡음뿐아니라 움직임이나 계기의 잡음은 fMRl의 데이터 분석을더욱 어렵게 한다. 따라서 실제 뇌신경의 활성화를 정확히 나타내는 참고데이터(reference data)를 선택하는 것은 힘든 일이며, 뇌신경의 활성화를 반영하는 의미 있는 여러 신호 형태에 대한 분석은 현재 fMRl의 후처리 (post-processing) 분석 방법에서 하나의 연구 과제라 할 수 있다. 본 연구에서는 prioriknow­-ledge 혹은 참고 데이터가 필요 없는 분석 방법인 Independent Component Analysis (lCA) 를 이용하여 fMRI선호를 분석하였다. ICA는 현재 많이 사용되고 있는 상관 분석 방법에 비해 신호의 형태를 분석하는 데에 보다 효과적일 수 있으며, 지연된 반응 형태를 갖는 신호나 움직임에 의한 신호의 패턴을 분리하여 분석할 수 있다. 한편, ICA만으후 fMRl의 신호에 따라 분석이 효과적이지 못한 경우 Principal Component Analysis(PCA) threshold, wavelet spatial f filtering, 부분적 영상 분석 방법들을 ICA전에 수행 함으로써 보다 효과적인 분석을 수행할 수 있다. ICA는 fMRl 신호의 형태 분석에 효과적인 방법이라고 생각하며, 데이터의 자유도를 감소 하기 위해서는 선 필터링 (pre-filtering) 방법들이 적용될 수 있다.

The fMRI signals are composed of many various signals. It is very difficult to find the accurate parameter for the model of fMRI signal containing only neural activity, though we may estimating the signal patterns by the modeling of several signal components. Besides the nose by the physiologic motion, the motion of object and noise of MR instruments make it more difficult to analyze signals of fMRI. Therefore, it is not easy to select an accurate reference data that can accurately reflect neural activity, and the method of an analysis of various signal patterns containing the information of neural activity is an issue of the post-processing methods for fMRI. In the present study, fMRI data was analyzed with the Independent Component Analysis(ICA) method that doesn't need a priori-knowledge or reference data. ICA can be more effective over the analytic method using cross-correlation analysis and can separate the signal patterns of the signals with delayed response or motion related components. The Principal component Analysis (PCA) threshold, wavelet spatial filtering and analysis of a part of whole images can be used for the reduction of the freedom of data before ICA analysis, and these preceding analyses may be useful for a more effective analysis. As a result, ICA method will be effective for the degree of freedom of the data.

We often extract a new feature from the original features for the purpose of reducing the dimensions of feature space and better classification. In this paper, we show feature extraction method based on independent component analysis can be used for classification. Entropy and mutual information are used for the selection of ordered features. Performance of classification based on independent component analysis is compared with principal component analysis for three real data sets.

We applied the ICA method to separate the ventricle and tissue components and to extract left ventricular input function from the H$_2$$^{15}$ O myocardial PET under the assumption that the elementary activities of ventricular pools and myocardium were spatially independent, and that the mixture of them composed dynamic PET frames. ICA-generated left ventricular input functions were compared with the ROI-generated ones, and also with the invasively derived arterial blood samples. Moreover, the rMBF calculated with the ICA-generated input functions and single compartment model was correlated with the results obtained with the radiolabeled microspheres.

In general, Independent component analysis (ICA) is a statistical blind source separation technique, used either in spatial or temporal domain. The spatial or temporal ICAs are designed to extract maximally independent sources in respective domains. The underlying sources for spatiotemporal data (sequence of images) can not always be guaranteed to be independent, therefore spatial ICA extracts the maximally independent spatial sources, deteriorating the temporal sources and vice versa. For such data types, spatiotemporal ICA tries to create a balance by simultaneous optimization in both the domains. However, the spatiotemporal ICA suffers the problem of source ambiguity. Recently, constrained ICA (c-ICA) has been proposed which incorporates a priori information to extract the desired source. In this study, we have extended the c-ICA for better analysis of spatiotemporal data. The proposed algorithm, i.e., constrained spatiotemporal ICA (constrained st-ICA), tries to find the desired independent sources in spatial and temporal domains with no source ambiguity. The performance of the proposed algorithm is tested against the conventional spatial and temporal ICAs using simulated data. Furthermore, its performance for the real spatiotemporal data, functional magnetic resonance images (fMRI), is compared with the SPM (conventional fMRI data analysis tool). The functional maps obtained with the proposed algorithm reveal more activity as compared to SPM.

In this research, for the first time, we tried to analyse Raman hyperspectral dentin data using Independent Component Analysis (ICA) to see its possibility of adoption for the dental analysis software. We captured hyperspectral dentin data on 569 spots on a molar with dental lesion by HR800 Micro Raman Spectrometer at UMKC-CRISP (University of Missouri at Kansas City-Center for Research on Interfacial Structure and Properties). Each spot has 1,005 hyperspectral data. We applied ICA to the captured hyperspectral data of dentin for evaluating ICA approach, and compared it with the well known multivariate analysis method, PCA. As a result of the experiment, ICA approach shows better local characteristic of dentin than the result of PCA. We confirmed that ICA also could be a good method along with PCA in the dental analysis software.

Long-term potentiation (LTP) of synaptic transmission is the most widely studied model for learning and memory. However its mechanisms are not clearly elucidated and are a subject for intense investigation. Previous attempts to decipher cellular mechanisms and network properties involved a current-source density analysis (CSDA) of the LTP from small animal hippocampus measured with a limited number of microelectrodes (typically <3), only revealing limited nature of spatiotemporal dynamics. Recent advancement in multi-electrode array (MEA) technology allows continuous and simultaneous recordings of LTP with more than 60 electrodes. However CSDA via the standard Laplacian transform is still limited due to its relatively high sensitivity toward noise, inability of resolving overlapped current sources and sinks, and its requirement for tissue conductivity values. In this study, we propose a new methodology for improved CSDA. Independent component analysis and its joint use (i.e., Joint-ICA) are applied to extract spatiotemporal components of LTP. The results show that ICA and Joint-ICA are capable of extracting independent spatiotemporal components of LTP generators. The ICs of LTP indicate the reversing roles of current sources and sinks which are associated with LTP.

Satellite and airborne hyperspectral sensor images are suitable for investigating the vegetation state in agricultural land. However, image data obtained by an optical sensor inevitably includes mixels caused by high altitude observation. Therefore, mixel analysis method, which estimates both the pure spectra and the coverage of endmembers simultaneously, is required in order to distinguish the qualitative spectral changes due to the chlorophyll quantity or crop variety, from the quantitative coverage change. In this paper, we apply our agricultural independent component analysis (ICA) model to an airborne hyperspectral sensor image, which includes noise and fluctuation of coverage, and estimate pure spectra and the mixture ratio of crop and soil in agricultural land simultaneously.

In order to extract only the alpha activity related signals from EEG recordings, we have applied Constrained Independent Component Analysis (cICA), a new extension of ICA in which some a priori knowledge of the alpha activity is utilized to extract only desired components. Its extraction (or filtering) performance has been compared to that of the conventional band-pass filtering via the scalp alpha power maps and cortical source maps of the alpha activity. Our results demonstrate that the alpha power maps and cortical source maps from the cICA-extracted alpha signals reveal more focalized alpha generating regions of the brain than those from the band-pass filtered alpha EEG signals. Furthermore they match more closely the activated regions of the brain mapped using fMRI, validating our results. We believe that the cICA-based filtering approach of EEG signals is a more effective means of extracting a specific brain activity reflected in EEG signals that will result in more accurate source localization or imaging maps.

Facial expression recognition is an intensive research area for designing Human Computer Interfaces. In this work, we present a new facial expression recognition system utilizing Enhanced Independent Component Analysis (EICA) for feature extraction and discrete Hidden Markov Model (HMM) for recognition. Our proposed approach for the first time deals with sequential images of emotion-specific facial data analyzed with EICA and recognized with HMM. Performance of our proposed system has been compared to the conventional approaches where Principal and Independent Component Analysis are utilized for feature extraction. Our preliminary results show that our proposed algorithm produces improved recognition rates in comparison to previous works.

For estimating the original signals only through observing the mixed data, independent component analysis (ICA) is a useful dimension reduction method, since it contains more statistical concepts: independence and non-Gaussianity. In other words, it does not end up in solving (generalized) eigenvalue problems, which are mostly restricted in considering up to the second moment as many other dimension reduction methods are. In this paper, we reviewed and explored various methods of ICA, such as Fast-ICA, Infomax-ICA, joint approximate diagonalization of eigenmatrices ICA (JADE ICA), product density estimation ICA (ProDenICA), and distance covariance ICA (dCovICA). We also proposed a method based on ball covariance, called B-dCovICA. Compared to other methods, B-dCovICA showed relatively high performance, supported by the higher accuracy of results when applied to simulated/real data. B-dCovICA is better than dCovICA, in that a non-parametric way is used to measure dependence, yet showed higher accuracy than dCovICA.

We address the problem of Blind Source Separation(BSS) of superimposed signals in situations where one signal has constant or slowly varying intensities at some consecutive locations and at the corresponding locations the other signal has highly varying intensities. Independent Component Analysis(ICA) is a major technique for Blind Source Separation and the existing ICA algorithms fail to estimate the original intensities in the stated situation. We combine the advantages of existing sparse methods and Kernel ICA in our technique, by proposing wavelet packet based sparse decomposition of signals prior to the application of Kernel ICA. Simulations and experimental results illustrate the effectiveness and accuracy of the proposed approach. The approach is general in the way that it can be tailored and applied to a wide range of BSS problems concerning one-dimensional signals and images(two-dimensional signals).