Redução de ruído em sensores de monitoramento usando separação cega de fontes

• Ano de defesa: 2016
• Autor(a) principal: Santos, Daniel Moraes
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Uberlândia; Brasil; Programa de Pós-graduação em Engenharia Elétrica
• Palavras-chave em Português: Separação cega de fonte; Análise estatística; Sensores; Monitoramento; Análise de componentes independentes; Ruído; Previsibilidade temporal; Taxa máxima sinal ruído; Blind source separation; Statistical analysis; Sensors; Monitoring; Independent component analysis; Noise; Temporal predictability; Maximum signal noise ratio; Engenharia elétrica; Detectores; Ruído elétrico; CNPQ::ENGENHARIAS::ENGENHARIA ELETRICA
• Link de acesso: https://repositorio.ufu.br/handle/123456789/18072; http://doi.org/10.14393/ufu.te.2016.151

Blind Source Separation (BSS) is known to be an efficient and powerful process to separate and estimate individual mutually independent signals acquired by various types of monitoring sensors. Theses monitoring sensors capture signals that are composed of various types of sources, the desired sources, unwanted sources and noisy sources. Thus, the desired signal is compromised so that it can be analyzed, this can lead to inefficient decision making. Ideally, the analyzed signals should be composed of the higher level of desired sources, and lower level of unwanted sources and noisy sources. This paper proposes an algorithm to identify and reduce noise in monitoring sensor signals using Blind Source Separation. This algorithm can be applied in any area of monitoring. It can identify noise without any kind of previous information of the signal analyzed. Initially, the algorithm makes the separation of the signals that were acquired by the sensors. These signals may have suffered influence from several noise sources. Different from the standard BSS, which requires at least two sources, this algorithm removes the noise from each signal separately applying the Maximum Signal-to-Noise Ratio and Temporal Predictability algorithms. The proposed algorithm also produces two outputs for each signal, the estimated original signal and the estimated noise. The results satisfy all the proposed objectives of this work. All the metrics used as parameters to evaluate the results obtained by the proposed algorithm were satisfactory. Specifically, for the thermal profile data, the most interesting results were the thermal gradients and their respective standard deviations, which showed a significant improvement