Classification of different magnetic structures from image data using deep neural networks

Abstract

We apply machine learning, specially deep neural network approaches, to train a new model that can perform an effective classification of ferromagnetic, anti ferromagnetic, skyrmion, anti-skyrmion and spin spiral configurations via supervised learning and also observe how the pre trained models like VGG16, VGG19, ResNet, Inception behave while solving this problem, draw a pattern from it and suggest path for further improving the model. The problem relies in categorization of Mag netic Configurations amongst many from input samples of simulation data in order to retrieve classified outcome from several different magnetic configurations. The data for the input sample derives from simulations of physical properties of various magnetic configurations. First CNN is used to classify between the images. Image classifications are mostly carried out using neural networks where data is placed in a graphical structure. In addition, the SVM method is applied twice, once with PCA and once without PCA. The proposed model in this research paper can success fully classify amongst magnetic configurations in real time with data obtained from spin-polarized scanning tunneling and Lorentz transmission electron microscopy. In our approach, we used a single deep neural network architecture is classify all five types of magnetic structures. All in all, this is a holistic approach for solving the classification problem of magnetic configuration and taking a step into optimizing the model.