Under global warming, the frequent occurrence of Eurasian extreme cold events in the Northern Hemisphere in recent years has attracted widespread attention. Winter Eurasia near-surface anticyclone activities are closely related to low temperatures. This study uses a new objective anticyclone identification algorithm based on a Mask region-based convolutional neural network (Mask R-CNN) to detect long-term winter Eurasian anticyclone activities. We found that most tracks of Eurasian anticyclones (30.5%) pass Siberia in winter. Anticyclone tracks over Siberia can be well separated into the eastward moving type, the northwest-southeast type, and the local type, with distinct associated circulation patterns. The energy dispersions of upstream Rossby waves have an important influence on the movement, maintenance and intensification of the three types. In association with anticyclones intensifying in Siberia, surface air temperatures in the mid-latitudes of Eurasia, especially Siberia, are significantly lower. In particular, the local type is the most active (56.1%) and is significantly correlated with the winter temperature anomaly in most areas of East Asia. Further research indicates that the preceding autumn sea ice condition in the Barents Sea can be used as a precursor factor to modulate the local type anticyclone activity and winter temperature anomaly in East Asia. In years of less ice, a quasi-meridional wave activity flux originating from the Barents Sea is more active, and the centre SLP of the local type anticyclone is higher with more cold air mass carried, which in turn results in colder weather in Siberia, the Mongolian Plateau and East Asia.