Research And Implementation Of Radiation Source Individual Identification Technology Based On Deep Learnin

In the field of target reconnaissance,how to realize the accurate identification of individual emitter has always been a key research issue.Compared with the traditional emitter recognition method,deep learning is widely used to improve the emitter signal recognition performance with its strong self-learning ability.However,when applied in complex electromagnetic environment,non-ideal working environment and data distribution will lead to low robustness and low recognition rate of the model.This paper aims at the problem of emitter individual identification under the influence of noise interference,small sample restriction,dynamic channel,and so on.It introduces deep residual shrinkage network,generation of confrontation network,and deep sub-domain adaptive network to carry out research on signal processing,model design and experimental verification,and builds a emitter identification system based on deep learning for experimental verification.The main research contents and innovation points are as follows:1.In order to solve the problem that signal noise affects the accuracy of emitter individual identification,a emitter individual identification method based on deep residual shrinkage network is proposed.This method uses the deep learning method and the soft threshold function to obtain a set of thresholds for soft thresholding,thus improving the ability to extract the signal features in the noise,and avoiding the problem that the threshold is difficult to select in the traditional wavelet denoising method.After that,the attention mechanism is automatically adjusted to extract important features and suppress irrelevant features while weakening the impact of noise,thus effectively improving the classification and recognition accuracy.The experimental results show that the overall recognition rate of the emitter classification recognition model proposed in this paper reaches 98.2% at 0d B,which is superior to other deep learning network models such as Resnet and improved Alexnet,and effectively reduces the impact of noise on the recognition accuracy.2.In order to solve the problem that small sample data sets affect the accuracy of emitter individual identification,a emitter individual identification method based on the generation of countermeasures network is proposed.This method is based on the deep convolution generation antagonism network,and introduces the self-attention mechanism in the discriminator and generator to improve the integrity and authenticity of the generated samples.In addition,an improved deep convolution neural network is designed.Compared with the traditional convolution neural network,the overall structure of the network is deeper and thinner,while still improving the recognition accuracy.The experimental results show that when the signal-to-noise ratio is 0d B and the original sample number of each individual is 40,the recognition accuracy is improved by 23.6% after using the proposed method to double the data.Compared with DCGAN-DCNN method,the recognition accuracy of this method is improved by 3.5%.3.In order to solve the problem that the inconsistent distribution of training set and test set affects the accuracy of emitter individual identification,a deep sub-domain adaptive emitter individual identification method is proposed.This method models the dynamic channel recognition problem as a domain adaptive problem from the perspective of migration learning.On the basis of the traditional deep learning model,the local maximum mean difference adaptive component is added to align the homogeneous sub-domains,and the model recognition performance is improved by extracting the fine information of the close-range ion category.The experimental results show that when the signal-to-noise ratio of the test signal increases or decreases by 4d B,the recognition accuracy of the method described in this paper is 11.4% and 12.7% higher than that of the domain-free adaptive method Resnet-50,respectively.4.In order to promote some key breakthrough technologies to engineering application,a emitter individual identification system based on deep learning method is built.In the algorithm structure,the complex neural network is designed according to the IQ characteristics of the emitter signal,and the residual attention mechanism is further introduced to enhance the extraction ability of fine features while extracting the nonlinear characteristics of the emitter.The hardware platform uses the interphone equipment as the signal source and USRP-2974 as the signal receiver.The received signal is analyzed and processed after being down-converted to baseband twice.In terms of software design,the system has the functions of signal acquisition and storage,receiving parameter setting,waveform display and classification recognition.The measured results show that in the actual communication process,the overall recognition rate reaches 99.7% under the natural environment noise.