Research On Special Emitter Identification Based On Machine Learning

The special Emitter Identification is one of the core technology in the electronic reconnaissance system,and it is also the forefront of research in the field of electronic warfare at home and abroad.It measures the characteristics of the received electromagnetic signal,matches the electromagnetic characteristics of the radiation source with the individual radiation source,and then identifies the type of equipment that emits the electromagnetic signal.The special emitter identification results play a key role in analyzing the communication network structure,determining enemy threat levels,and tactical decisions.Facing the complex and changeable electromagnetic environment,the traditional radiation source identification method has been unable to meet the requirements of the battlefield environment for identification performance and application scope.With the development of artificial intelligence,new breakthroughs and progress have been made in the use of deep learning for radiation source identification.Therefore,this paper comprehensively uses deep learning methods to carry out the following work from the three aspects of the overall scheme design of individual source identification,model performance optimization research,and source identification under small sample conditions:1.A convolutional neural network recognition algorithm based on deep timefrequency feature extraction is proposed.This algorithm uses self-encoder and convolutional neural network to achieve fine fingerprint feature extraction,and solves the problem of significantly reduced recognition performance when the number of radiation source individuals increases or the radiation source signal becomes complicated.The simulations are performed using real-source radiation source signals.The results show that when the number of individual radiation sources increases from 3 to 5,6,9 and the complexity of the signals gradually increases,the algorithm's recognition accuracy is always stable above 95%.It is obviously better than the traditional recognition methods,which proves the accuracy and robustness of the algorithm in this paper,and effectively meets the needs of indicators for practical applications.2.An algorithm for adaptively adjusting the learning rate based on a loss function is proposed.This algorithm uses the current loss function value and the previous loss function value to adjust the learning rate step size factor without the need for frequent manual parameter adjustments.It solves the problems of slow learning speed,timeconsuming training,and overcoming the optimal solution caused by the difficulty of setting the learning rate parameters in the neural network.The simulations are performed using real source signal,and the results show that under the same conditions,compared with the existing learning rate,the learning rate algorithm based on the loss function proposed in this paper effectively improves the convergence speed and accuracy of the individual source recognition model The number of training iterations is reduced and the model performance is optimized.3.A weighted transfer extreme learning machine algorithm is proposed.The algorithm uses a case-based transfer learning and extreme learning machine theory to implement a "knowledge transfer" from the source domain to the target domain under a small sample condition.Solved the problem that in the actual tactical communication network,due to the insufficient number of labeled samples of the radiation source,the individual identification of the newly added radiation source was difficult.The simulations were performed using real-world radiation source signals.The results show that the algorithm proposed in this paper effectively improves the accuracy of individual identification of newly added radiation sources under small sample conditions,which is an average increase of 3% over the DAELM algorithm and an average increase of 29% over the traditional ELM algorithm.,Which proves the effectiveness of the algorithm.