To address the issue of insufficient anti-jamming capabilities caused by intelligent and multi-means adversarial jamming in future complex battlefield electromagnetic environments, a dynamic adaptive reliable transmission waveform decision-making method based on reinforcement learning is proposed. Building on the reinforcement learning algorithm, the greedy action selection within the algorithm is first optimized to balance the exploration probability during the early and late stages of the algorithm. Additionally, a self-attention mechanism is incorporated to enable the algorithm to capture relationships between different inputs. The optimized algorithm is used to establish an intelligent agent, which is then trained to make decisions on anti-jamming transmission methods. Simulation results show that, compared to traditional reinforcement learning algorithms, the optimized algorithm achieves a 10% higher average reward value in the early exploration phase and converges 20% faster on average during the mid-phase. Finally, a bit error rate performance analysis of the transmission methods output by the decision-making process verifies the effectiveness of the waveform decision-making strategy.