Research On Radar Unknown Target Recognition Of High Resolution Range Profile Technology Based On Deep Learning

In the automatic target identification based on High Resolution Range Profile(HRRP),it is usually necessary to obtain target data,and establish a corresponding feature library after training the model to perform target recognition.In practical applications,because the samples of certain targets cannot be obtained in advance,the correct recognition results cannot be given.Therefore,it is necessary to identify unknown targets before performing target recognition.Compared with the traditional shallow learning model,the deep learning model can acquire the deeper essential characteristics of the target and is more robust to target recognition.Therefore,this article uses deep learning to identify unknown targets.The content of this article is as follows:1.A method of Deep Convolutional Network + Probability Difference Discriminant Sub-network to identify unknown targets is proposed.This method uses the translation invariance of the Deep Convolutional Neural Network to effectively extract the known target features,and then inputs them into the discriminant sub-network to identify unknown targets.The Probability Difference Discriminant Sub-network can describe the probability that the target belongs to each known category,so as to more accurately discriminate the unknown target.Experimental results show that the method has a recognition rate of 85.78% for unknown targets when the signal-to-noise ratio is 10 dB.Compared with the recognition method based on Gaussian kernel,it can stably recognize unknown targets.2.An unknown target recognition method based on Deep Belief Network + Nearest Neighbor Discriminating Sub-network is proposed.Using the label database of known targets and the Deep Belief Network(DBN),mining the high-dimensional features of the target data,and then input the feature data into the Nearest Neighbor Discriminating Sub-network for recognition.The DBN network can better mine the depth features of the High Resolution Range Profile data,and the discriminant sub-network uses the nearest neighbor value to compare the similarity between the target and the data in the library,and combines the KPCA algorithm to achieve dimensionality and noise reduction,so as to more accurately identify the unknown target.Simulation experiments show that the recognition method using the Deep Belief Network is 36.11% higher than the SVDD method based on the Gaussian kernel under 5dB noise,and it has better noise immunity.3.Research an unknown target recognition model based on stacked long short-term memory network(SLSTM)+ nearest neighbor discriminant sub-network.The SLSTM network can use the relevant information between adjacent samples to extract more target features,and then use the nearest neighbor discriminating sub-network to identify unknown targets.The results of simulation experiments show that the recognition rate of unknown targets using SLSTM method is stable up to 96% under different target conditions to be recognized.At the same time,in the case of changes in noise,compared with traditional machine learning methods,the method based on SLSTM + nearest neighbors to discriminate sub-networks also reflects better performance.