Research On Radar Emitter Signal Recognition Method Based On Improved Cooperative Semi-supervised Learning

Radar radiation source identification as an important part of modern electronic warfare,after the receiving device intercepts the enemy radar radiation source signal,it is necessary to further analyze the signal to obtain the radar radiation source signal type and other information.In recent decades,with the continuous updating of radar technology,various new radars have been used on the battlefield,leading to the increasingly complex electromagnetic environment faced by the modern battlefield.Using the traditional method of manually extracting and selecting features,it will face problems such as slow identification speed,low identification accuracy,and difficult identification at a low signal-to-noise ratio,which is difficult to adapt to the rapidly changing situation of the modern battlefield.Therefore,how to accurately and quickly identify radar radiation source signals is an urgent problem in the current radar radiation source identification field.In recent years,with the continuous breakthrough of deep learning in various important fields,how to effectively combine deep learning and radar radiation source identification to solve the shortcomings and shortcomings of traditional methods has become one of the mainstream research directions of radar radiation source identification.To solve the problems such as poor recognition performance of existing radiation source recognition methods and difficult to recognize under harsh conditions.This paper introduces Temporal Convolutional Network(TCN)in natural language processing into radar radiation source identification,and proposes a fast identification method based on radar radiation source sequence data.First,the One-Dimensional Convolutional Neural Networks(1DCNN)is connected with the Temporal Convolutional Network(TCN),and the Re LU activation function in the network is replaced by the Leaky Re LU activation function.Meanwhile,a batch normalization layer is added before the One-Dimensional Convolutional Neural Network to accelerate the convergence,and an attention layer is added before the fully connected layer to filter the signals.After the simulation experiments using eight common radar radiation source sequence data,it can be seen that the model in this paper not only has a significantly higher recognition speed and higher recognition accuracy than other common models,which is a good solution to the problem that it is difficult to balance the recognition speed and recognition accuracy of other models.However,the above method works well provided that the available samples of radar radiation source signals are large and the model can be adequately trained.With the continuous progress of science and technology,the modern battlefield electromagnetic environment becomes more and more complex,the available samples of unknown radiation source signals will become less,and it is time-consuming and laborious to label the radiation source signals.If trained directly with the received radiation source signal samples,it will lead to overfitting of the model and affect the recognition performance.To solve this problem,this paper further proposes an improved cooperative semi-supervised classification algorithm capable of acting directly on a small amount of labeled and a large amount of unlabeled radiation source sequence data.By adding homogenization to the loss function to reduce the risk of over-absoluteness in classification,the recognition accuracy of semi-supervised learning is effectively improved.Moreover,to address the problem of long training time of conventional semi-supervised learning methods,the dynamic time warping technique is applied to further optimize the unlabeled data set,which reduces the training time and ensures the recognition accuracy of the network.Through simulation experiments,it is proved that the semi-supervised learning method adopted in this paper can solve the problem of overfitting due to the small number of labeled radar radiation sources,effectively improve the recognition accuracy of the network,and significantly increase the recognition speed compared with the common semi-supervised learning networks.