This study presents a long short-term memory (LSTM) neural network for classifying BPs in the solar chromosphere and transition region as oscillatory or nonoscillatory, including damped oscillatory BPs using the Interface Region Imaging Spectrograph (IRIS). The LSTM network, which has 500 hidden units, was trained using the full spectral profiles of the Mg II h line at 2796.354 Å, the Mg II k line at 2803.531 Å, and the Si IV line at 1394 Å. These lines comprise both the blue shifted and redshifted components. Integrating spectral diagnostics with manually annotated datasets allowed the model to achieve an average classification accuracy of 86% on a test set of 20 000 BPs held out from 20 IRIS datasets. Wavelet analysis and a damping model estimated decay times ranging from 101 to 265 s and maximum Doppler velocities ranging from 20 to 175 kilometers per second. A key finding was the 25–40 s temporal lag in the damping of Si IV oscillations compared to Mg II and between active region (AR) and coronal hole (CH). These lags suggest different wave-plasma interactions in network and internetwork areas. Simultaneous 1400 Å SJIs ensured damping within BPs. Swarm charts revealed regional variations, showing higher velocities in the AR internetwork (175 km/s) than in the CH (49 km/s). These findings impose further constraints on energy dissipation and wave propagation processes, enriching our understanding of solar atmospheric dynamics through deep learning.
Yuan et al. (Sun,) studied this question.