A Underwater Target Recognition Technique Based On Support Vector Machine

This dissertation is carried on one of National defense pre-research project.As the "brain" of the underwater platform,the fuze is responsible to recognize and judge the targets around the underwater platform based on receiving their radiating signal.The basic requirement of the fuze was classifying the targets rapidly and accurately,which is a research hotspot in intelligent decision system.Support vector machine(SVM)is a better pattern classification solution in this field and the preferred classifier in recent years.Based on the small sample statistical theory and the structural risk minimization principle,SVM is strong stabile,efficient and robust.According to practical application of fuze,this dissertation summarizes the theory and algorithm of SVMs,and had a try on classifier experiments to the latest ships based on the features of targets radiating signal which were selected by kernel principal component analysis,the classification results is satisfactory.In the beginning,the application value of support vector machine(SVM)in underwater platform classification is analyzed.Then the advantages of SVM in a limited sample identification model is summarized,which is based on describing the core idea of statistics theory,the basic principle of SVM and the different construction methods for classification in detail.Secondly,analyzing the basic characteristic of the ships radiated noise,the elements of joint feature were extracted,such as the type of the noise,the through and spectrum characteristic,the characteristics of autocorrelation curve in time domain,Welch-1(1/2)dimensional spectrum in frequency domain and Wigner-1(1/2)spectrum in time-frequency domain.Next,the kernel principal component analysis(KPCA)method is studied to reduce the dimension of the joint feature vector.DDK-KPCA algorithm is implemented for selecting the joint feature vector based on specific backgrounds,which improves the efficiency of algorithm so that the computing burden of underwater platform hardware system is reduced.Fourth,The DDK-CK-SVM model had be validated by the target’s radiated noise in the laboratory,which is constructed using Composite Kernel based on Data Dependant Kernel(DDK-CK),and the average classification rate reached more than 185%.Compared with BP-NN classifier,the DDK-CK-SVM classifier is higher in both classification rate and efficiency.At the same time,combining with the development of mission requirements,the author designed and made the radiation noise analog signal preprocessing device.After testing in lab and field,this device has been applied to practical engineering projects.In conclusion,the model and algorithm in this dissertation had simulated and validated,and the classification result is satisfactory.The research is valuable to engineering practice.