Embedded system implementation for material recognition using deep learning

Material recognition is the process of classifying materials into different categories based on the constituent material of the object under study. It is a very important problem in many fields and especially in the industrial field. In factories that use production lines to manufacture their products, one step is to separate materials and package different materials. In that case, it helps to move each material into specific containers. This step can be automated, which saves money, time and can improve efficiency. We utilized the advances in deep learning to build a system for material recognition, and we tested it on the Flicker Materials Database (FMD) with an accuracy of 79.25%. We also built a local database using materials from our environment via a camera mounted on an Raspberry Pi 3 embedded system, and that gave an accuracy of 90.5%. The system is self-contained and can be portable in any factory with minimal changes.