Classification of Microstructure Images of Metals Using Transfer Learning

Mohammed Abdul Hafeez Khan, Hrishikesh Sabnis, J. Angel Arul Jothi, J. Kanishkha, A. M. Deva Prasad

Abstract: This research focuses on the application of computer vision to the field of material science. Deep learning (DL) is revolutionizing the field of computer vision by achieving state-of-the-art results for various vision tasks. The objective of this work is to study the performance of deep transfer learned models for the classification of microstructure images. With light optical microscopes, microstructure images of four different metals were acquired for this task, including copper, mild steel, aluminum, and stainless steel. The proposed work employs transfer learned powerful pre-trained convolutional neural network (CNN) models namely VGG16, VGG19, ResNet50, DenseNet121, DenseNet169 and DenseNet201 to train and classify the images in the acquired dataset into different classes of metals. The results showed that the transfer learned ResNet50 model has obtained the highest accuracy of 99%, outperforming other transfer learning models. This also shows that DL models can be used for automatic metal classification using microstructure images.