Study On Radar Target Recognition Using High Resolution Range Profiles

With the wide application of radar in military and civil affairs, the ability of target recognition has become increasingly demanding. Therefore, radar target recognition is one of the hot research interests and attention is focused on target recognition based on high resolution range profile(HRRP) in this paper. One-dimensional HRRP contains much structural information of target scattering distribution, which provides a lot of effective features for target classification. Time-frequency analysis is an important tool to study the relationship between signal frequency and time-varying, which is also an important feature extraction method in target recognition. Manifold learning aims to reveal the structural relationships among samples and has broad application prospects in feature extraction and dimensionality reduction.In this thesis, we mainly study different feature extraction methods used for target recognition based on high resolution range profile. The main contents of this thesis are summarized as follows:1. The method of target length extraction using HRRP is studied. On the basis of analysis and comparison of several existing length extraction methods, we present the adaptive threshold algorithm and bi-directional sliding average algorithm. Then the validity of the algorithms is verified via experimental data.2. The method of time-frequency feature extraction and target classification based on SVM are studied. Two-dimensional time-frequency diagram based on HRRP for target recognition can maximize the information utilization and help to improve the recognition performance. However, it contains lots of redundant information, which makes it essential to select the features with best discriminability in the time-frequency diagram. SVM can select the optimal features automatically by integrating the feature selection with classification. The experimental results indicate that the frequency characteristics selected by SVM are valid.3. The range profile feature extraction method based on manifold learning is studied. Numerous classical manifold learning algorithms are analyzed. And the SKOLPP is applied to the high resolution range profile target recognition. This algorithm adds the class label information of samples to LPP and utilizes the kernel function to generalize LPP to nonlinear form. In addition, the obtained projection features are mutual y orthogonal. The experimental results show that the algorithm has excellent performance.4. The target recognition method based on statistical geometrical features is studied. Several features that could reflect a certain amount of geometric structure information of the targets are extracted from a HRRP and used for recognition, including the overall entropy, the statistical size, the number of intense scatterer, the variance-mean ratio, the maximum minimum energy ratio, the entropy of intense scatterer. Furthermore, in order to make full use of the advantages of each feature, the features are combined with each other in a certain way to form new combination features for target classification. The experimental results indicate that the recognition performance based on combination features is much better than that based on the single feature.