Research On Robust Narrowband Target Classification Method Against Radar Parameters

The micro-Doppler features of target reflect its unique geometric characteristics,which is one of the important ways to realize target classification.The micro-moving parts of helicopter,propeller aircraft and jet aircraft have obvious differences in structure and motion characteristics,so the micro-Doppler modulation characteristics in the narrow band echo are also different.By extracting relevant micro-Doppler features,the classification of three aircraft targets can be realized.At present,most of the related researches assume that the radar working parameters are fixed in the process of micro-Doppler target classification.However,in the radar system working environment,in order to improve the anti-jamming performance and solve the problem such as blind speed,the radar carrier frequency and pulse repetition frequency will change,which will lead to the performance degradation or even failure of the classification algorithm.Aiming at this problem,the effects of the changes in these two radar parameters on the target micro-Doppler modulation are studied and the corresponding robust classification methods are proposed in this thesis.The main work of this thesis is summarized as follows:1.First,the echo models of horizontal and vertical rotors are established.Second,simulation experiments are carried out by using the models to analyze the difference of the microDoppler modulation between three types of aircraft targets,and the influence of blade angle and attitude angles are also analyzed.Third,the existing three types of aircraft target classification algorithm is introduced,and some specific implementation steps in preprocessing,feature extraction and classifier design are also introduced.2.The effect of change in radar carrier frequency is studied and it is pointed out that the spectral width and the number of spectral lines of the micro-Doppler modulation spectrum will change accordingly,and the spectral line spacing is less affected.The first method is based on the spectral interval,which is only related to the rotor’s rotation speed.It extracts the frequency-domain correlation features that are not sensitive to the carrier frequency variation to realize classification.Experiments demonstrate that this method realizes robust classification.The second method is based on neural network model and transfer learning,which makes full use of the similarity of training and test samples.Firstly,the neural network classification model is pre-trained with training samples,and then a small number of new samples of which carrier frequency is different from training samples are used to fine-tune part of the network parameters to realize classification.Experimental results show that this method can improve the robustness of classification.3.In order to solve the problem that the structure of micro-Doppler spectrum is changed due to the change of pulse repetition frequency in the process of classification,a classification method based on resampling preprocessing is proposed,which can improve the classification performance in the scene of pulse repetition frequency changed when the test sample spectrum is not overlapped.Then,according to the situation of spectrum aliasing of test samples,a classification method based on spatial pyramid pooling network is proposed.This method takes advantage of the fact that spatial pyramid pooling is not sensitive to the change of input signal dimension,and extracts features from multiple scales.The experimental results show that this method improves the classification performance under the condition of pulse repetition frequency variation.