Uso de autoencoders para remoção de ruído em imagens termográficas

Abstract

Non-destructive testing is one of the tools that allows the assessment of the proper functioning of equipment, as well as the quality of the materials used in them. It allows the monitoring of components without causing any damage to them. Among the various techniques used in this process, is infrared thermography, which allows the mapping of a body based on heat distribution. In this work, processing of thermographic images was carried out in order to treat possible noise that arises during the stage of obtaining the thermograms. To perform this task, an autoencoder based on a CNN was used, which has the role of collecting the attributes of the image and reconstructing it without the noise. It was proven that neural networks have a great potential to reduce noise in images, but their architecture must be more robust so that there is no great loss of information in the images.