부산대학교 도서관

In order to solve the problems of traditional radar fixed waveform parameters to perform tasks that are vulnerable to reconnaissance, this paper combines deep Q-learning networks in deep reinforcement learning with a posteriori estimation of the target state and proposes a cognitive radar waveform adaptive selection model based on an improved Deep Q-leaning Network (DQN) strategy. The proposed method models the radar environment, designs a reward function based on the target entropy state, and uses the DQN to learn the reward expectation for the target to select different waveform parameter actions in the current state. The experiment results show that the improved DQN policy-based cognitive radar waveform adaptive selection scheme proposed in this paper achieves significant efficiency gains and is more practical than traditional heuristic algorithms.