Identification and Classification of Aluminum Scrap Grades Based on the Resnet18 Model

In order to reduce the elemental species produced in recycling and melting of aluminum scrap improve quality pure alloys, it is necessary classify different grades before melting. For problem classifying scrap, most existing studies are conducted using laser-induced breakdown spectroscopy for identification classification, which requires a clean flat metal surface enormous equipment costs. this study, we propose new classification method based on ResNet18 network model, improves efficiency reduces cost. The objects research three scrap: 1060, 5052, 6061. features were compared machine vision algorithm; datasets, RGB, HSV, LBP, built comparison find best training dataset subsequent hyperparameters learning rate batch size tuned model. results show that there was differentiation threshold between through features; model trained showed RGB dataset. With hyperparameter optimization accuracy final recognition could reach 100% effectively achieve scrap.