目的 雷达辐射源识别是指先提取信号特征,再将特征输入分类器进行识别。随着电子科技水平的提高,各种干扰技术应用于雷达辐射源信号中,使得信号个体特征差异越来越不明显。仅靠传统的模板匹配,分类器设计,决策匹配等辐射源识别技术难以提取信号可辨性特征。针对这一问题,利用深度学习良好的数据解析能力,提出了一种基于卷积神经网络的辐射源识别方法。方法 根据雷达辐射源信号的特点,首先对未知辐射源信号进行传统域特征(频域,功率谱,信号包络,模糊函数代表性切片)提取。其次,从传统域特征内获得有效的训练样本特征集合,利用卷积神经网络自动获取训练样本深层个体特征得到辐射源识别模型。最后,将该模型应用于所有测试样本中,获得辐射源识别结果。结果 实验内容主要包括测试在不同域特征下卷积神经网络的识别性能,并将本方法与传统基于机器学习和最新基于深度强化学习中的深度Q网络模型(DQN)识别算法进行了对比,结果表明,当卷积神经网络的输入为频域特征时,本文提出的方法识别准确率达100%,相较于支持向量机(SVM)本文的识别准确率提高了0.9%,当输入为模糊函数切片特征和频域时,本文所提出的平均识别准确率相比SVM模型提高了16.13%,相比于极限学习机(ELM)分类器提高了1.87%,与当前流行的识别算法DQN相比,提高了0.15%。结论 实验结果表明本文提出的方法能有效提高雷达辐射源信号识别准确率。
objective In complex electronic warfare, radar emitter identification is an essential component of electronic intelligence and electronic support systems; its related technology remains a critical factor in measuring the level of electronic countermeasures equipment technology. Radar emitter identification mainly refers to extracting signal characteristics and then inputting the features into the classifier for identification. With the improvement level of electronic technology, various jamming techniques are applied to radar, which makes it hard to identify individual signal differences. It is challenging to extract signal discriminability characteristics by traditional radiation source identification techniques, such as template matching, classifier design, decision matching and so on. Given the robust data analysis capabilities of deep learning, convolutional neural networks(CNN) are one of the earliest and most widely used deep learning models. In general, CNN consists of the convolutional layer, the pooling layer, the activation layer, and the fully connected layer. The convolution layer and layer stack structure extract powerful features and different features of data, respectively. The activation layer is used to enhance the feature expression ability of the network. Pooling layers can reduce dimensions. Feature combination and classification are carried out at the full connection layer. According to the characteristics of the radar radiation source signal, we propose a new method of radar source identification based on convolutional neural networks. Methods In order to extract the individual implicit features of the radiation source signal, our CNN contains five layers; the first three are convolutional and the remain two are fully-connected. The kernel of the third convolutional layer and the pooling layer are set to one-dimensional. The rectified linear units(ReLU) non-linearity is applied to the output of every convolutional and fully-connected layer. We use dropout in the first two fully-connected layers which can combat overfitting. The first convolutional layer filters with 36 kernels of size 3 with a stride of 1. The second convolution layer consistent with the parameters of the first convolutional layer. The third convolutional layer has 64 kernels of size 5 with a stride of 1. The specific steps of the algorithm are as follows. Firstly, the original radar data is preprocessed, i.e., signal noise reduction and normalization. Secondly, we extract different signal characteristics of the signal. Finally, the convolutional neural network is trained using different features. Results The training set is 20%, 40%, 60%, 80% of the total number of samples, respectively. This article compares the recognition accuracy of convolutional neural networks when the input is a different feature and compares the recognition accuracy of support vector machine(SVM) model, extreme learning machine(ELM) model and depth Q network model(DQN) in deep reinforcement learning. Experiments show that when the training set ratio is 80%, the input of the network is different domain characteristics, a higher recognition rate can be obtained. Recognition accuracy rate reaches 100% and 99% for the spectral feature and fuzzy function slice feature respectively. When the input is the frequency domain feature with 80% of the training set, we compared the performance of support vector machine (SVM). Our method outperforms the SVM. Specifically, it improves accuracy by 0.9%. When the input is the fuzzy function slice feature, compared to SVM classifier and ELM classifier, the accuracy rate of our method is improved by 16.13% and 1.87% respectively. Compared with the current popular recognition algorithm DQN, it improves by 0.15%. Conclusions The convolutional neural network designed in this article has a strong feature expression ability. The experiment results show that our proposed extracts the implicit features of signal discrimination. It can improve the recognition accuracy of radar emitters.