Radar target recognition is an important direction of modern radar technology.It can provide the attribute,category and type of a target by the radar echo data.It also has an important influence of information and intelligent military equipment for the future.In the light of the high-resolution range profile obtained by radar,using the target recognition algorithm to classify and identify research.the support vector data description with the negative class sample method and multi-core support vector data description based on hybrid kernel function method is studied,and make improvement on this basis,then validate and test the effectiveness of the algorithm by using the simulation data and actual one-dimensional high-resolution range profile.First,chapter one expounds the background and significance of the whole research,briefly illustrates the system framework of radar automatic target recognition(RATR)technology and development situation at home and abroad.The target identification technology on high resolution range profile is summarized,at the same time,the research status of the support vector data description(SVDD)algorithm is introduced.Then explained the relationship of the support vector machine algorithm and support vector data description algorithm,and teased out the principle of the two algorithms in detail,and analyzes the influence of SVDD recognition results when the main parameters of kernel function are changed,then verified by experiments.Secondly,the multinuclear SVDD-neg target identification method based on hybrid kernel function is studied.The nature of the kernel function and the Mercer conditions are introduced respectively,and analyzes the nature of local kernel function and global kernel function,a multiple kernel SVDD-neg algorithm based on hybrid kernel function is proposed.This method has the advantages by combine the generalization ability of polynomial kernel function and the learning ability of gaussian kernel function,constructs the new hybrid kernel function to solve the traditional SVDD algorithm the kernel function of the single led to limited model recognition performance effect.Then comparing the method with classical SVDD recognition algorithm,and analysis the performance of algorithm through the experiment.Finally,based on the multi-kernel SVDD algorithm theory,the introduction of incremental learning method,multiple-kernel ISVDD algorithm is given.Firstly,the impact of multi-kernel SVDD is analyzed when new incremental sample joined theoretically.Then multi-kernel ISVDD algorithm and the algorithm flow is given based on the thought of incremental learning,and analyze the algorithm by the experimental.This method can guarantee algorithm recognition rate,at the same time,can reduce training time of incremental sample in some degree. |