Dynamic spectrum control is a viable anti-jamming scheme; however, frequency switching (FS) has relatively high time overhead in practice. Moreover, if the jamming variation cycle is shorter than that of the FS, it can cause communication degradation or even failure. Therefore, we employ both link layer FS and low-overhead physical layer actions for joint anti-jamming and establish distinct decision timescales based on these overheads. Specifically, we design distinct networks for different actions, utilizing cognitive information as rules to guide the agent optimization direction for resource deployment. Simulation results indicate that our method improves the throughput by 112.87% compared to the same timescale algorithm, which demonstrates the effectiveness of associating cross-layer actions at different timescales under highly dynamic interference conditions.