Examinando por Materia "SMV"

Mostrando 1 - 2 de 2
  • Miniatura
    PublicaciónAcceso abierto
    Algoritmo para el reconocimiento de las características de la señal electroencefalográfica bajo anestesia general
    (Universidad EIA, 2013) Gómez Suárez, Nicolás Eugenio; Torres Villa, Róbinson Alberto
    El presente trabajo busca implementar un algoritmo para reconocer las características de la señal electroencefalográfica (EEG) de pacientes bajo anestesia general. Las características que se buscan reconocer son: la Entropía Aproximada (ApEn), la Entropía de Shannon o de la Información, el valor eficaz (RMS) y el centroide.
  • Miniatura
    PublicaciónAcceso abierto
    Desarrollo de un sistema de detección de emociones mediante el procesamiento de señales fisiológicas
    (Universidad EIA, 2021) Carrasquilla Quintero, Andrés; Mejía Mejía, Elisa
    ABSTRACT: Emotions play an important role on mental processes, such as cognition, perception, learning, communication, and decision-making. Therefore, it is important for devices to be capable of having knowledge of the user’s emotional state, without the need of providing explicit feedback, in order to both monitor the subject’s mental health, and design good user experiences. There are several available databases containing physiological signals associated with emotional stimuli. One such database (DEAP) was chosen using selection criteria. The signals were processed using Matlab to extract 22508 features, and an iterative process of feature selection was executed considering the Fisher score, the area under the ROC curve, and the correlation coefficients. Three classification algorithms were developed with different data: features from all physiological signals, features from approximately half of the subjects, and features extracted from EEG. In each case, the best results were obtained when using support vector machines (SVM). The greatest accuracy rate achieved was 79.46% for valence, and 76.16% for arousal. All three systems had a notably high performance. The development of these algorithms was carried using a Subject Independent (SI) approach, that is, one classification algorithm for all of the subjects. This results in a more generalizable method than most classification algorithms reported in the literature, as they tend to use a Subject Dependent (SD) approach. The classification systems that have been proposed in this work are comparable to many of these algorithms, and even have a better performance than some DS systems, that often produce better results since they had been designed for a particular subject. Future possible approaches to improve the developed systems were proposed, such as having an algorithm with multiple layers of classification, using physiological signals from more subjects, or subdividing each signal into smaller pieces to process them individually and generate more features.