The eXtremely Large-scale Multiple-Input-Multiple-Output （XL-MIMO） system can significantly improve channel capacity. However， traditional Uniform Linear Arrays （ULAs） experience a drastic reduction in the near-field region at large incident and emitted angles， leading to limited signal coverage. The use of Uniform Circular Arrays （UCAs） can effectively expand the near-field regions， but renders low-overhead beam training schemes based on ULA impractical. To reduce the overhead of near-field beam training with UCA， a new fast beam training scheme was proposed. In the first stage， UCA was approximated as ULA， and a joint method was used to construct a far-field hierarchical codebook for angle domain user search； in the second stage， based on the angles obtained from the first stage， UCA was used for exhaustive search in both angle and distance domains. Simulation results on a UCA system with 512 antennas indicate that the proposed scheme requires only 28 training overheads， while maintaining good robustness across different Signal-to-Noise Ratio （SNR） conditions， and its rate performance achieves 99.16% of the benchmark.