Study On Radar HRRP Target Recognition

A high-resolution range profile (HRRP) is the amplitude of the coherent summations of the complex time returns from target scatterers in each range resolution cell, which represents the projection of the complex returned echoes from the target scattering centers onto the radar line-of-sight (LOS). It contains the target structure signatures, such as target size, scatterer distribution, etc., thereby radar HRRP target recognition has received intensive attention from the radar automatic target recognition (RATR) community. The theory and techniques for radar HRRP target recognition are researched from the three aspects, i.e. analysis on the physical property of HRRP samples, feature extraction and feature selection, and classification methods, in this dissertation, which are supported by Advanced Defense Research Programs of China (No. 413070501 and No. 51307060601) and National Science Foundation of China (No. 60302009).The main content of this dissertation is summarized as follows.·Based on the scattering center model, the first part makes a detailed analysis on the physical property of HRRP samples, and points out that how to deal with the target-aspect, time-shift and amplitude-scale sensitivity of HRRP samples is a challenging task for radar HRRP target recognition. Then a framework for HRRP-based RATR by template matching method (TMM) is established, which forms the basis for the following study.·The second part focuses on feature extraction and feature selection for HRRP-based RATR. The main work includes: 1) Due to the huge storage requirement and the complex computation, higher-order spectra features receive less attention from RATR community. Similar to the well-known kernel method in automatic target recognition (ATR) community, in which the Euclidean distances in the high dimensional mapped space can be calculated in the low dimensional original space, a method for calculating the Euclidean distances in higher-order spectra feature space is proposed in this dissertation, which is performed in original HRRP space directly and can avoid calculating the higher-order spectra, effectively reducing the computation complexity and storage requirement. 2) According to the scattering center model, a new feature extraction method using the amplitude fluctuation property of HRRP samples is proposed in this dissertation. The weighted HRRP vector extracted by the new method can effectively fuse the corresponding frame's stcatterer strength distributing profile and variance profile, and represent the scatterer distribution in every range cell, thereby it can describe the scattering property of a target better. 3) Based on the Fisher's linear discriminant, a weighted feature selection method is proposed. According to the characteristics of radar HRRP target recognition, the proposed weighted feature selection method use the iterative algorithm based on the Fisher's discriminant ratio to search the optimal weight for the time-shift invariant feature, i.e. power spectrum. Compared with using the raw feature vectors and some existing feature selection methods, the weighted feature selection method not only can reduce the feature dimension, but also can improve the recognition performance with low computational complexity.·The third part is contributed to radar HRRP statistical modeling. The main work concerns the following three aspects. Firstly, we make a detailed analysis on the effect of the three sensitivity problems of HRRP samples on statistical recognition, and propose our corresponding solution, which forms the basis for the study on radar HRRP statistical modeling. Secondly, under the hypothesis that the elements in an HRRP sample are statistically independent, we develop an independent statistical model comprising two distribution forms, i.e. Gamma distribution and Gaussian mixture distribution, to model echoes of different types of range cells as different distribution forms. Thirdly, theoretical analysis and our experimental results based on measured data show that the independence assumption is not true, thus we further make a study on the more accurate statistical recognition methods based on HRRP samples' jointly statistical characteristics. Our work includes: 1) Different from general target recognition problems, L₂ normalized samples are applied to HRRP-based RATR to deal with the amplitude-scale sensitivity problem, therefore, geometrically speaking, HRRP samples spread on a unit hypersphere. We propose a modified statistical recognition method based on subspace approximation for power transformed HRRP samples under the joint-Gaussian distribution hypothesis. 2) According to the experiments based on measured data, HRRP samples approximately follow the joint-Gaussian distribution described by factor analysis (FA) model. Therefore, we can apply FA model to radar HRRP statistical recognition rather than a joint-Gaussian mixture model, e.g. FA mixture model, which is a more accurate choice for modeling non-Gaussian distributed correlations in multidimensional data but with high learning complexity and large computation burden. Furthermore, an iterated algorithm for model selection of FA model in radar HRRP statistical recognition is proposed, which can automatically give the optimal aspect-frame boundaries and determine the optimal number of factors in each aspect-frame.·The forth part focuses on RATR using complex HRRP samples. Based on the analysis on the physical property of complex HRRP samples, we point out that the frame template, classification algorithm and feature extraction method for complex HRRP samples should be unvaried with the initial phases. In the existing classification methods, the principal component analysis (PCA)-based minimum reconstruction error approximation is independent of the initial phases yet exploits the remaining phase information in complex HRRP samples, therefore, this method can be used in complex HRRP-based RATR, and a fast time-shift compensation algorithm is proposed for this method. In addition, we propose a novel feature extraction method invariant with the initial phases for complex HRRP samples. The recognition algorithms, frame-template establishment methods and preprocessing methods used in the real HRRP-based RATR can also be applied to the proposed complex feature vector-based RATR. Therefore, based on the aforementioned research, complex HRRP-based RATR becomes practical. Furthermore, in the recognition experiments based on measured data, complex HRRP samples can obtain better recognition results than real HRRP samples.·The fifth part is contributed to multicategory classification by a small number of simple classifiers. Due to the target-aspect sensitivity of HRRP samples, radar HRRP target recognition is a typical multicategory classification problem. We propose a multicategory classification method based on hypercube/hypergrid self-organization mapping (SOM) scheme. The advantageous of this method includes: 1) We can not only use binary classifiers but also k-ary (k<