一种深度强化学习的雷达辐射源个体识别方法

针对传统依赖于人工经验提取辐射源个体特征的不足，提出一种基于深度强化学习的雷达辐射源个体识别方法。利用发射机非理想信道造成的辐射源信号包络在信号变化时呈现的不同瞬态信息，以信号包络前沿作为深度神经网络的输入状态，以辐射源类别作为当前输入状态的可选动作，通过卷积神经网络自动提取辐射源包络个体特征，并拟合当前状态动作对的Q值，进而以强化学习模型完成雷达辐射源个体识别任务。讨论了深度Q网络模型、深度双Q网络模型以及Dueling Network模型3种深度强化学习模型在辐射源识别任务中的应用。实测数据仿真实验表明：传统机器学习算法的识别率不足80%，而深度强化学习网络的识别率高达98.42%.

Specific Emitter Identification Based on Deep Reinforcement Learning

A specific emitter identification (SEI) method based on deep reinforcement learning is proposed on account of the deficiency of emitter individual feature extraction depending on artificial experience. Due to the differences of the transient information of signal envelope, which results from the change of the signal owing to a nonideal transmitter channel, an envelope rising edge is used as the input state of deep neural network, and the emitter classifications are used as the optional actions of the current input state. The envelope features are extracted automatically through the convolutional neural network (CNN), and Q values of the current state action pairs are fitted, thus completing the specific emitter identification task based on the reinforcement learning model. The applications of deep Q network (DQN), deep double Q network (DDQN) and Dueling network in the specific emitter identification are discussed. The measured results show that the recognition rate of traditional machine learning algorithm is less than 80%, but the deep reinforcement learning model can achieve the high recognition rate of 98.42%.