Analysis of periodic structures with Fourier description and neuronal network

Abstract

This work is developed in a project of textile lattices inspection. The structure of a superficial texture is manifested with a behavior on the base of a model known as pattern which, is associated with a group of characteristics that define it as such. The identification process and classification of shortcomings in the texture consists on establishing a region of conformity in the coordinated space defined by the pattern’s characteristics. The reduction process of this m-dimensional space, helps to its identification in an n-dimensional space, such that m > n, where the classification system actually depends on the characteristics of the new space, where the new characteristics truly contain the classification information. The space characteristics allow for the identification of the pattern as such in the place that is explored. The characteristic frequency corresponds to a reduction of the classification space, making it more generic in the area over the image. The classification system is modeled with neuronal networks algorithms and the complexity of the surfaces of decision is solved starting from the architecture and the algorithms of training of the neuronal net.