Such As Target Recognition Method Based On The Distance Of One-dimensional Kernel Function Radar

The technology of radar target recognition, which plays an important role in modern radar system and is widely and intensively focused by all of the countries, has been one of the key components of the present and future defense weapon system. The range profiles of targets, containing abundant information for classification and recognition, can be obtained easily by high-range resolution radar. Many methods based on the kernel function for feature extraction and pattern classification are adopted to analyze and study radar target recognition by range profiles intensively and extensively in this dissertation. The main work of this dissertation is listed as follows:1.According to the algorithm of kernel fisher discriminant analysis (KFDA), a concept, named kernel sample, is introduced. Based on this concept, KFDA is equivalent to performing the algorithm of linear fisher discriminant analysis (LDA) on kernel sample sets. This dissertation adopts the algorithm of principal component analysis (PCA) to properly reduce dimension of kernel samples. Then, these kernel samples reduced dimension are utilized to perform LDA, when performing LDA , the within class and between class scatter matrixes are diagonalized simultaneously to extract optimal features of kernel samples. Above-mentioned scheme is applied to radar range profile's recognition, experimental results show that it can give comparatively accurate recognition rate in range of tiny aspect angle of radar targets.2.The algorithm of support vector machines (SVM) is introduced to try to completely solve radar range profile's recognition in range of huge aspect angle of targets. Based on standard SVM on unclassifiable sample sets condition (C-SVM), a kind of generalized and modified C-SVM algorithm is introduced into further improving on effect of radar range profile's recognition, furthermore, the generalized C-SVM algorithm can still commendably dispose an extreme status of radar target recognition. At the same time, some effectual training samples, which relate to pattern classification, are filtrated, via the algorithm presented by this dissertation, to reduce complicated degree of SVM algorithm, shorten study time and economize resource. The experimental results show that above-mentioned algorithms are valid and feasible. 3.The algorithm of least squares support vector machines (LS-SVM) is introduced to try to completely solve radar range profile's recognition in range of huge aspect angle of targets. Based on standard LS-SVM, a kind of improved LS-SVM algorithm named generalized and modified LS-SVM algorithm is presented by this dissertation to further improve on effect of radar range profile's recognition. Recognition rate of the existing LS-SVM sparse algorithms rapidly decreases with the reduction training samples in dealing with some pattern recognition problems including radar range profile's recognition. So, a new LS-SVM sparse algorithm is proposed in order to overcome this difficulty, improve performance efficiency with the recognition sample via LS-SVM algorithm. The inverse matrix of high dimension is easily solved through association of iterative increment LS-SVM and the new sparse algorithm united, which is propitious to hardware implementation. The experimental results show that above-mentioned methods are valid and feasible.