Evaluation of Classifiers for a Vision-based Automated Mold Protection System Using Modified LBP Features

For decades, plastic components have been the main parts of products in industries such as food, pharmaceutical, automotive, etc. Generally, these components are created by injection molding machines. Using these machines, raw materials are converted to plastic parts, e.g., bottle caps, dosing spoons, and bumpers. The part of the machine that provisionally holds plastic products is called “Mold” which has a unique form for each product. Since molds are sensitive components with high prices, appropriate care is required. When mold is used as the dynamic part of the machine, it’s a high potential for damages due to incomplete product ejection. Utilizing an automated inspection system is a modern solution to prevent possible problems. In this paper, we propose an intelligent system based on machine vision that consists of image capturing, processing, and classification sections. In the processing section, we have used a novel modified Local Binary Pattern algorithm which leads to the suitable features for classifying images into two categories. To achieve the best classifier, four potent machine learning-based methods are evaluated: KNN, SVM, Random Forest, and Gradient Boosting. This evaluation is based on criteria like F1-score, training and processing time, and the experimental results claim that the SVM method is the best classifier with 11.87ms training time, 9.04us processing time, and F1-Score of 0.96.