Time-resolved magneto-optical imaging assisted by deep-learning video frame interpolation
Time-resolved magneto-optical imaging enables us to observe spatio-temporal propagation dynamics of spin waves, collective precession motion of magnetic moments propagating as waves. Combining with Fourier-based analysis of the images, the dispersion relation of the excited spin waves in magnetic materials can be reconstructed. Since the dispersion relation modulation is inevitable to realize various spin-wave applications such as wave-computing devices, spin-wave dispersion measurement with high frequency and wavenumber resolution is expected. The typical energy scale of the interactions that modulate spin-wave dispersion is in MHz range, which is comparable to the highest limit of the frequency resolution of current techniques such as Brillouin light scattering and Neutron spectroscopy. We report deep-learning assisted spin-wave dispersion measurement method to realize observation of spin-wave dispersion relation with a high-frequency resolution. We use a convolutional neural network (CNN) to interpolate frames between two different images in the proposed method. We improved interpolation quality by taking the wave nature of the image into account in the training, which leads to application of video interpolation technique for spin-wave spectroscopy.