Visão computacional aplicada ao controle de qualidade de casulos do bicho-da-seda: subsistema de classificação de imagens

Abstract

The quality control in the silk fiber industry is traditionally done in manual fashion. With the objective to automate the visual inspection selection of adequate cocoons in the silk production, it is proposed a computer vision system to substitute visual inspection done, as example, in the BRATAC company production line. This work’s objective is the development of a pilot algorithm for the silkworm cocoons image classification subsystem. To the development and evaluation of this selection algorithm, three image databases of adequate and defective cocoons were formed; texture, shape and SIFT descriptors extraction algorithms were developed and descriptor similarity was evaluated in its ability to classify cocoons in a group adequate to silk production and a defective group. The algorithms were developed in Matlab platform and promising results were shown to to the silkworm cocoon classification.