Unmanned aerial vehicles have gradually emerged as a core force in missions such as reconnaissance and strike operations, as well as earthquake relief, thanks to their advantages of high efficiency and low cost. However, UAV networks are exposed to external interference during mission execution, and the strength of their anti-interference capability is directly related to the success or failure of the missions. To address the problem of UAV channel access in a dynamic interference environment, this paper integrates problem solving with a centralized training and distributed execution framework, and proposes a multi-agent UAV dynamic channel decision-making algorithm based on the value decomposition network. In addition, the fixed learning rate in deep reinforcement learning is improved to a cosine annealing learning rate, thereby achieving lower losses and higher model accuracy. Simulation results show that the proposed algorithm enables network nodes to intelligently adjust channel selection strategies and maximize the throughput of wireless networks, while the cosine annealing algorithm enhances the training effect of reinforcement learning.