Research on multiwavelength isolated bright points based on deep learning

Multiwavelength bright points (BPs) are taken to be cross sections of magnetic flux tubes extending from the surface of the photosphere upward to the higher photosphere. We aim to study the characteristics of isolated multiwavelength BPs using the cotemporal and cospatial TiO band and Hα line wings from the Goode Solar Telescope at Big Bear Solar Observatory. A deep-learning method, based on Track Region-based Convolutional Neural Networks, is proposed to detect, segment, and match the BPs across multiple wavelength observations, including the TiO, Hα + 1 A, Hα - 1 A, Hα + 0.8 A, and Hα - 0.8 A line wings. Based on the efficient detection and matching result with a precision of 0.98, 1283 groups of BPs matched in all five wavelengths are selected for statistics analysis. The characteristic values of the BPs observed at the same red and blue line wings are averaged. For the BPs of the TiO, averaged Hα ± 1 A, and averaged Hα ± 0.8 A line wings, the mean equivalent diameters are 162 ± 32, 254 ± 33, and 284 ± 28 km, respectively. The maximum intensity contrasts are 1.11 ± 0.09, 1.05 ± 0.03, and 1.05 ± 0.02aIQSn, respectively. The mean eccentricities are 0.65 ± 0.14, 0.63 ± 0.11, and 0.65 ± 0.11, respectively. Moreover, the characteristic ratios of each Hα ± 1 A and Hα ± 0.8 A BP to its corresponding TiO BP are derived. Hα ± 1 A and Hα ± 0.8 A line wings BPs show 60% and 80% increases compared to TiO BPs, respectively. With increasing height, most BPs almost keep their shapes. This work is helpful for modeling the three-dimensional structure of flux tubes.