Context. The study of magnetic activity in the Sun's polar regions is essential for understanding the solar cycle. However, measuring polar magnetic fields presents challenges due to projection effects and their intrinsically weak magnetic field strength. Faculae, bright regions on the visible solar surface associated with increased magnetic activity, offer a valuable proxy for measuring polar fields.

Aims. This research aims to analyze the magnetic activity of the Sun's polar regions through the use of polar faculae.

Methods. A neural network model (U-Net) was employed to detect polar faculae in images from the Helioseismic and Magnetic Imager (HMI) on board the Solar Dynamics Observatory (SDO). The model was trained on synthetic data, eliminating the need for manual labeling, and was used to analyze 14 years of data from May 2010 to May 2024.

Results. The U-Net model demonstrates superior performance and efficiency over existing methods, enabling automated large-scale studies. We find that polar faculae numbers exhibit cyclical behavior with distinct minima and maxima, showing similar patterns between poles but with notable temporal delays (south pole: minimum early 2014, maximum late 2016; north pole: minimum late 2014, maximum mid-2019). Polar faculae magnetic fields remain consistent in magnitude (∼±75 G) across both poles and throughout the solar cycle. A strong linear correlation was found between the polar faculae count and the overall polar magnetic field strength. The spatio-temporal evolution reveals systematic migration of field polarity reversals from mid-latitudes toward the poles at rates of 3−8 m/s. During solar minimum, we observe a small relative increase in stronger-field faculae compared to solar maximum, suggesting either the coexistence of two magnetic distributions or subtle solar cycle dependence in faculae properties.