Support Vector Machines Learning Algorithm And Its Application In Radar Target Recognition

Support Vector Machines (SVM) is a new pattern recognition method derived from Statistical Learning Theory, which has many adventages in pattern recognition, such as its superiority in small-sample, nonlinear and high-dimension problems. SVM resolves the shortcomings of neural networks and other traditional classification methods effectively for its good performance and high generalization ability. Therefore, SVM will be promising for its application in radar target recognition.However, as a new technique, SVM method still has some shortcomings, such as the training speed is very slow for large-scale data, not supporting incremental learning, the problem of multi-class SVM, etc. These severely limit the widespread application of SVM.SVM learning algorithms are researched and applied in radar target recognition based on HRRP in this paper. The mainly works can be shown as follows:1. A fast training algorithm for SVM based on K nearest neighbours is proposed. The new algorithm extracts border vectors which may be support vectors and trains SVM by substituting the border vectors set for training set. Experimental results show the algorithm enhances the speed of training SVMs greatly, while the ability of classification can be guaranteed.2. A fast incremental learning algorithm for SVM based on K nearest neighbours is presented. The algorithm extracts border vectors set and trains SVM by substituting the border vectors set for training set. The method reduces training samples largely and advances training speed greatly, while the ability of SVM to classify is guaranteed because the border vectors set contains all useful training samples. The experiment result shows the effectiveness of the algorithm.3. A SVM multiclass classification algorithm based on kernel hierarchical clustering is proposed, which not only enhances the speed of training SVM effectively, but also acquires higher precision and speed for classification. Experimental results indicate that the algorithm has higher generalization ability.4. The algorithms proposed in the paper are applied in radar target recognition based on HRRP and the effects are better.