Abstract
Computer-aided materials recognition from images is critical for automation in materials industry. In this manuscript, we perform machine learning studies to recognize common materials from images. To this end, an annotated materials image database consisting of typical materials including metal, plastic, glass, fabric, leather and wood is provided. Subsequently, a PSPNET-based deep learning model is constructed to classify these materials from images. The model achieves decent recognition accuracies above 0.60 for leather and wood, which are comparable to human. The machine learning process is also attempted for videos and small-size images to further demonstrate the viability of the image-based machine learning techniques. This study highlights the importance of image-based deep learning studies for materials science, and calls for further machine learning studies to automate and expedite the materials recognition process.
Subject
General Physics and Astronomy