A Novel Thermal Image based Metal Classification System using Machine Learning Algorithms

• Published in: IEEE access Vol. 13; p. 1
• Main Authors: Chiluveru, Siva Rajesh, Rana, Shuvendu
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 01.01.2025; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Accuracy; Algorithms; Aluminum; Classification; Copper; Cost effectiveness; Decision Tree; Decision trees; Heating systems; Iron; Machine learning; Magnetic induction; Magnetic resonance imaging; Magnetic separation; Metal Classification; Metals; Quality control; Random Forest; Recycling; Specific heat; Thermal conductivity; Thermal Images; Thermal imaging; X-ray fluorescence
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2025.3610992

Classifying metals is an essential task in all industries to make sure the materials used in the processes are safe and meet the required standards all while enhancing operational and cost effectiveness. Metal classification is crucial in processes such as quality control, recycling, green manufacturing and any other processes that rely on metal. The metallurgical domain relies on techniques for metal classification such as X-ray Fluorescence, Magnetic Induction Spectroscopy and Eddy Current Separation with regards to their speed, accuracy and cost. This work proposes a new system for thermal image based metal classification which is expected to be better in terms of effectiveness, cost and time as compared to the traditional methods. In this work, heat conductivity and specific heat characteristics of metals are taken into account to generate a thermal image map. As different metals have different conductivity and specific heat, the heat absorption and radiation map generated by thermal image can be used to identify the type of metal using a machine learning based classification method. The results show that fused with machine learning Decision Tree and Random Forest algorithms in the metal identification system can classify metals with an accuracy of 96% and 98% respectively. These results highlight how machine learning and thermal imaging can offer more precise and reliable results than traditional machinery. This approach presents a practical solution that conserves cost and time.