Deep learning approach for image classification of magnetic phases in chiral magnets

Chiral magnets have been vastly studied in the last years due to their potential applications as information carrying devices. In this work, it was studied the influence of the magnetocrystalline anisotropy term on the magnetic properties of two-dimensional chiral magnets, by means of Monte Carlo simulations. On the obtained magnetic arrangements, a Deep Learning algorithm based on computer vision techniques, was used for the phase recognition in each sample and their later classification. In this way, the low-temperature phase diagrams were built for a given set of anisotropy constants. It was observed that the skyrmion phase can be stabilized by tuning the magnetocrystalline anisotropy constant; in particular, the skyrmion region in the low-temperature magnetic phase diagrams can be increased by using an easy-plane anisotropy term. The machine learning algorithm described in this work can be easily extended to real-world applications, in the analysis and classification of magnetic force or Lorentz transmission electron microscopy images.