SISTEMA DE APRENDIZAJE BASADO EN VISIÓN POR COMPUTADOR PARA LA INSPECCIÓN AUTOMATIZADA DE PARTÍCULAS MAGNÉTICAS EN ESTRUCTURAS MARINAS

This work presents a system of supervised learning based on computer vision with the aim of solving the automation of non-destructive inspection tests based on magnetic particles. In this paper, three supervised learning algorithms have been tested: the nearest k neighbor (kNN), a Bayesian classifier (NBC) and the vector support machine (SVM). The developed system has been successfully tested on a set of images extracted during the inspection of magnetic particles on marine structures at the Navantia shipyard in Cartagena. The algorithm that offered the best result was the SVM with a sensitivity of 98.6% and a specificity of 100.0% in the detection of faults by magnetic particles. The vector of characteristics used is composed of a set of 16 elements formed by geometric characteristics and intensity values of the RGB, HSV, and CIE L * a * b * color spaces. The work presents a software application and a hardware system that, using the SVM algorithm, is capable of automatically detecting defects on marine structures during the magnetic particle test. Keywords. Magnetic particles, Non-destructive testing, Machine learning, Computer vision