Application of a Paraconsistent Analysis Network (PANnet) as a binary classification system of metallic alloys images based on the studies of parameters selection João Luís Lopes Freitas Orsi Kuntz, João Inácio Silva Filho

Abstract

 This paper describes the study of parameters selection and the process of creating a PANnet capable of performing the binary classification of images for the presence or absence of rust in images of metallic alloys. It is shown that the processing of Haralick’s texture feature descriptors by the structures provided by the Paraconsistent Logic (PL), with the threshold calculated via the Receiver Operating Characteristic Curve (ROC Curve), allows the creation of a classification system with 75% accuracy and a calculated f-score of 0.78.