Extração de Características do Padrão Speckle para Classificação de Perturbações em Fibra Óptica

• Ano de defesa: 2021
• Autor(a) principal: Reis, Ingrid Andrade
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal do Espírito Santo; BR; Mestrado em Engenharia Elétrica; Centro Tecnológico; UFES; Programa de Pós-Graduação em Engenharia Elétrica
• Palavras-chave em Português: Extração de características; análise de textura; speckle; LBP; GLCM; redes neurais; subject.br-rjbn; Engenharia Elétrica
• Link de acesso: http://repositorio.ufes.br/handle/10/15624

In multimode optical fibers, a speckle pattern, or specklegram, appears at the output of the fiber when illuminated by coherent light. In this case, the phenomenon responsible for generating the specklegram is the interference between the different modes propagating in the fiber. Considering the sensitivity of this pattern to changes in the optical fiber, sensors capable of detecting different types of disturbances, such as vibrations, stress and displacements, have been developed. Previous works show that there is a correlation between the distance at which a disturbance is generated in an optical fiber and the changes that occur in its speckle pattern. Due to its granular appearance, it is proposed in this work a speckle pattern image classification system using neural networks based on features extracted by texture descriptors, in order to assess whether such aspects can also represent the specklegram. For this, two datasets containing images obtained by experiments with polymeric optical fibers were used and, for each one, the results of accuracy for different sets of characteristics were compared. They were extracted by two texture extractors, the Local Binary Pattern (LBP) and the Gray Level Co-occurrence Matrix (GLCM). The results showed that it was possible to classify the location of the perturbations, especially when using the uniform and rotation-invariant LBP operator applied to the images when divided into 25 blocks.