Álvaro Orozco-Gutiérrez

Comparative analysis of physiological signals and electroencephalogram (EEG) for multimodal emotion recognition using generative models

Multimodal Emotion recognition (MER) is an application of machine learning were different biological signals are used in order to automatically classify a determined affective state. MER systems has been developed for different type of applications from psychological evaluation, anxiety assessment, human-machine interfaces and marketing. There are several spaces of classification proposed in the state of art for the emotion recognition task, the most known are discrete and dimensional spaces were the emotions are described in terms of some basic emotions and latent dimensions respectively. The use of dimensional spaces of classification allows a higher range of emotional states to be analyzed. The most common dimensional space used for this purpose is the Arousal/Valence space were emotions are described in terms of the intensity of the emotion that goes from inactive to active in the arousal dimension, and from unpleasant to pleasant in the valence dimension. The use of physiological signals and the EEG is well suited for emotion recognition due to the fact that an emotional states generates responses from different biological systems of the human body. Since the expression of an emotion is a dynamic process, we propose the use of generative models as Hidden Markov Models (HMM) to capture de dynamics of the signals for further classification of emotional states in terms of arousal and valence. For the development of this work an international database for emotion classification known as Dataset for Emotion Analysis using Physiological signals (DEAP) is used. The objective of this work is to determine which of the physiological and EEG signals brings more relevant information in the emotion recognition task, several experiments using HMMs from different signals and combinations of them are performed, and the results shows that some of those signals brings more discrimination between arousal and valence levels as the EEG and the Galvanic Skin Response (GSR) and the Heart rate (HR).

Feature extraction of speech signals in emotion identification

In this work, the acoustic and spectral characteristics and the automatic recognition of human emotional states through speech analysis have been studied. Acoustic features have been evaluated and features from time-frequency representation are proposed. The method is based in the representation of speech signal through energy distributions (Gabor transform and WVD) and discrete coefficients (DWT and linear prediction analysis). Recognition accuracy of 94.6% for emotion detection are obtained from SES database of emotional speech in spanish language.

SVM-based feature selection methods for emotion recognition from multimodal data

Multimodal emotion recognition is an emerging field within affective computing that, by simultaneously using different physiological signals, looks for evaluating an emotional state. Physiological signals such as electroencephalogram (EEG), temperature and electrocardiogram (ECG), to name a few, have been used to assess emotions like happiness, sadness or anger, or to assess levels of arousal or valence. Research efforts in this field so far have mainly focused on building pattern recognition systems with an emphasis on feature extraction and classifier design. A different set of features is extracted over each type of physiological signal, and then all these sets of features are combined, and used to feed a particular classifier. An important stage of a pattern recognition system that has received less attention within this literature is the feature selection stage. Feature selection is particularly useful for uncovering the discriminant abilities of particular physiological signals. The main objective of this paper is to study the discriminant power of different features associated to several physiological signals used for multimodal emotion recognition. To this end, we apply recursive feature elimination and margin-maximizing feature elimination over two well known multimodal databases, namely, DEAP and MAHNOB-HCI. Results show that EEG-related features show the highest discrimination ability. For the arousal index, EEG features are accompanied by Galvanic skin response features in achieving the highest discrimination power, whereas for the valence index, EEG features are accompanied by the heart rate features in achieving the highest discrimination power.

Método para el Diagnóstico de Rodamientos Utilizando la Complejidad de Lempel-Ziv

La presencia de una falla en un rodamiento hace que el sistema mecanico evolucione de una dinamica debilmente no lineal a una dinamica fuertemente no lineal, por lo tanto los metodos lineales comunes en el dominio del tiempo y la frecuencia no son adecuados para el diagnostico de rodamientos. En el presente articulo se propone una metodologia novedosa no lineal para la deteccion de fallas en rodamientos, que usa la medida de complejidad sugerida por Lempel y Ziv para caracterizar las senales de vibracion. La ventaja principal de este metodo sobre las demas tecnicas de analisis no lineal es que no requiere la reconstruccion de un atractor, por lo que es adecuado para realizar analisis en tiempo real. Los resultados obtenidos muestran que la complejidad de Lempel-Ziv es una herramienta efectiva para el diagnostico de rodamientos.

On detecting determinism and nonlinearity in microelectrode recording signals: Approach based on non-stationary surrogate data methods

Two new surrogate methods, the Small Shuffle Surrogate (SSS) and the Truncated Fourier Transform Surrogate (TFTS), have been proposed to study whether there are some kind of dynamics in irregular fluctuations and if so whether these dynamics are linear or not, even if this fluctuations are modulated by long term trends. This situation is theoretically incompatible with the assumption underlying previously proposed surrogate methods. We apply the SSS and TFTS methods to microelectrode recording (MER) signals from different brain areas, in order to acquire a deeper understanding of them. Through our methodology we conclude that the irregular fluctuations in MER signals possess some determinism.

Feature relevance estimation for vibration-based condition monitoring of an internal combustion engine

Condition monitoring of Internal Combustion Engines (ICE) benefits cost-effective operations in the modern industrial sector. Because of this, vibration signals are commonly monitored as part of a non-invasive approach to ICE analysis. However, vibration-based ICE monitoring poses a challenge due to the properties of this kind of signals. They are highly dynamic and non-stationary, let alone the diverse sources involved in the combustion process. In this paper, we propose a feature relevance estimation strategy for vibration-based ICE analysis. Our approach is divided into three main stages: signal decomposition using an Ensemble Empirical Mode Decomposition algorithm, multi-domain parameter estimation from time and frequency representations, and a supervised feature selection based on the Relief-F technique. Accordingly, we decomposed the vibration signals by using self-adaptive analysis to represent nonlinear and non-stationary time series. Afterwards, time and frequency-based parameters were calculated to code complex and/or non-stationary dynamics. Subsequently, we computed a relevance vector index to measure the contribution of each multi-domain feature to the discrimination of different fuel blend estimation/diagnosis categories for ICE. In particular, we worked with an ICE dataset collected from fuel blends under normal and fault scenarios at different engine speeds to test our approach. Our classification results presented nearly 98% of accuracy after using a k-Nearest Neighbors machine. They reveal the way our approach identifies a relevant subset of features for ICE condition monitoring. One of the benefits is the reduced number of parameters.

Kernel-Based Relevance Analysis with Enhanced Interpretability for Detection of Brain Activity Patterns

We introduce Enhanced Kernel-based Relevance Analysis (EKRA) that aims to support the automatic identification of brain activity patterns using electroencephalographic recordings. EKRA is a data-driven strategy that incorporates two kernel functions to take advantage of the available joint information, associating neural responses to a given stimulus condition. Regarding this, a Centered Kernel Alignment functional is adjusted to learning the linear projection that best discriminates the input feature set, optimizing the required free parameters automatically. Our approach is carried out in two scenarios: (i) feature selection by computing a relevance vector from extracted neural features to facilitating the physiological interpretation of a given brain activity task, and (ii) enhanced feature selection to perform an additional transformation of relevant features aiming to improve the overall identification accuracy. Accordingly, we provide an alternative feature relevance analysis strategy that allows improving the system performance while favoring the data interpretability. For the validation purpose, EKRA is tested in two well-known tasks of brain activity: motor imagery discrimination and epileptic seizure detection. The obtained results show that the EKRA approach estimates a relevant representation space extracted from the provided supervised information, emphasizing the salient input features. As a result, our proposal outperforms the state-of-the-art methods regarding brain activity discrimination accuracy with the benefit of enhanced physiological interpretation about the task at hand.

A comparison of robust Kalman filtering methods for artifact correction in heart rate variability analysis

Heart rate variability (HRV) has received considerable attention for many years, since it provides a quantitative marker for examining the sinus rhythm modulated by the autonomic nervous system (ANS). The ANS plays an important role in clinical and physiological fields. HRV analysis can be performed by computing several time and frequency domain measurements. However, the computation of such measurements can be affected by the presence of artifacts or ectopic beats in the electrocardiogram (ECG) recording. This is particularly true for ECG recordings from Holter monitors. The aim of this work was to study the performance of several robust Kalman filters for artifact correction in Inter-beat (RR) interval time series. For our experiments, two data sets were used: the first data set included 10 RR interval time series from a realistic RR interval time series generator. The second database contains 10 sets of RR interval series from five healthy patients and five patients suffering from congestive heart failure. The standard deviation of the RR interval was computed over the filtered signals. Results were compared with a state of the art processing software, showing similar values and behavior. In addition, the proposed methods offer satisfactory results in contrast to standard Kalman filtering.

EEG seizure identification by using optimized wavelet decomposition

A methodology for wavelet synthesis based on lifting scheme and genetic algorithms is presented. Often, the wavelet synthesis is addressed to solve the problem of choosing properly a wavelet function from an existing library, but which may be not specially designed to the application in hand. The task under consideration is the identification of epileptic seizures over electroencephalogram recordings. Although basic classifiers are employed, results rendered that the proposed methodology is successful in the considered study achieving similar classification rates that had been reported in literature.

Emotion Assessment Using Adaptive Learning-Based Relevance Analysis

The study of brain electrical activity (BEA) allows to describe and analyze the different cognitive and physiological process that occurs inside the human body. The Electroncephalogram (EEG) is often chosen over other neuroimaging techniques, but the non-stationarity nature of the EEG data and the variability between subjects have to be sorted to design reliable methodologies for neural activity identification. In this work, we propose the use of adaptive filtering for the relevance analysis of EEG segments in emotion assessment experiments. First, a windowing stage of the EEG data is performed, from which brain connectivity measures are extracted as BEA descriptors. The correlation and the time-series generalized measure of association (TGMA) are selected at this stage. Then, the connectivity data is used for galvanic skin response (GSR) and Blood Volume pressure (BVP) estimation employing the quantized kernel mean least squares (QKLMS) strategy. Finally, from the QKLMS algorithm, a set of relevant centroids in the estimation of physiological responses are used in the classification of the specific emotional state. The results obtained validate the proposed methodology and give clear evidence that a selection of segments from BEA improve further stages of classification for emotion assessment tasks.