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

Abstract

Along the silk production chain, silkworm’s cocoons pass through successive quality control steps, usually involving human visual inspections. However, the physical and psy chological fatigue caused by repetitive execution of such a task can impair workers’ health and the quality of the final product. In this context, the long-term objective of the project is to automate the whole process of detection, classification and removal of cocoons that don’t meet the quality criteria. The objective of this work, in turn, is to analyze the different existing image segmentation methods, focusing on the detection of silkworm’s cocoons. The methodology includes the generation of a cocoons’ image bank and the implementation and comparison of 3 different image segmentation methods, chosen by a comparison of the characteristics of the problem and the characteristics of the algorithms available in the literature. In the end, the results of the 3 segmentation algorithms are presented and a comparative analysis of their performance, advantages and disadvantages is made.