Research On Feature Extraction And Fusion Recognition Based On High-resolution One-dimensional Range Profile

Radar automatic target recognition is a key technology in modern radar information processing,and has been widely used in military and civilian field.The high resolution range profile(HRRP)obtained from wideband radar reflects the distribution of the target scattering centers along the radar line of sight,which contains enough information about the structure and shape of the target.Compared with two-dimensional or three-dimensional radar imagery,HRRP is easier to be captured.Besides,real-time recognition can be realized.Therefore,radar target recognition based on HRRP is of huge potentials.In this dissertation,aiming at the target of aircraft and taking high resolution range image as the research object,we study in depth the key issues as feature extraction,multiple features fusion,classifier design in target recognition both theoretically and experimentally.Below are the primary work of this dissertation.1.The first part study the characteristics of the radar HRRP.We discuss the radar scattering center model and one-dimensional HRRP acquisition method,and point out that how to deal with the target-aspect,time-shift and amplitude-scale sensitivity of HRRP simples.2.The second part is contributed to the feature extraction method.In addition to using classic subspace method to extract features,we also use deep neural network to extract features from HRRP.Through experiments,the classification ability of each feature is compared,and the characteristics of each feature are analyzed.3.The third part focus on the target recognition method of feature-level fusion.Based on the idea of genetic algorithm,different features are fused together,the population initialization strategy of traditional genetic algorithm is improved.In the process of evolution,the adaptive crossover probability and mutation probability are developed.Experimental results show that this method not only reduce the dimension of data,but also reserve the valid information on the greatest degree and can achieve better recognition performance.4.The fourth part focus on the target recognition method of decision-level fusion.Based on the idea of that the same classifier has different classification ability for different samples,an adaptive weighted voting fusion recognition model is designed byusing the posterior probability of the multi-class relevance vector machine.Experimental results indicate that this fusion strategy can effectively improve the recognition rate.5.A Monte Carlo fusion method for target recognition based on neural network is studied.Multiple feature of a target are fused by genetic algorithm firstly,and then the Monte Carlo fusion method is used to fuse the multiple neural networks.Experimental results demonstrate that this method can further improve the performance of system identification.