Otimização por enxame de partículas para segmentação de cordões de solda em imagens radiográficas de tubulações de petróleo

Abstract

The radiographic weld inspection is important to ensure quality and security of pipe networks. Despite all training and knowledge, specialists can provide misclassifications for several reasons. The visual tiredness, distraction and the quantity of radiographic to be analyzed can induce an inspector mistaken. This work aims at proposing an assistant system for automatic segmentation of weld bead present in radiographic images. For this, it presents an approach for segmentation of weld bead in radiographic images of type double wall double image (DWDI), merging two well known algorithms: Particle Swarm Optimization - PSO and Dynamic Time Warping - DTW. A vertical profile is raised from a window encoded in a particle of PSO and it is compared through DTW with a model profile. The similarity measure between the model and extracted profile is the basis for the fitness computation which is of great importance to the final performance. Thus a lot of effort has devoted to choose a suitable fitness function for PSO. The tests were realized in two steps: firstly, the height of the extracted profile was fixed and in a second set of experiments the height was a variable component incorporated into the coding of particles and evolved during PSO iterations. The results obtained in automatic segmentation of weld bead showed that the PSO, mostly, converged satisfactorily (first phase achieved a performance was between 85,17% and 93,11% and the second one is between 79,83% and 81,36%), to the window that enables the segmentation of the weld bead, indicating promising results.