Research On Algorithm Of SVM And Software Simulation On CCS

Statistical learning theory (SLT) and support vector machines (SVM) provide methods to solve the problems that samples are infinite. SLT is a theory specialized in machine learning with finite samples, and SLT provides a firm foundation to SVM. SVM can trade off between the complexity and learning abilities, so it has high generalization ability. It is considered as a candidate to replace neural networks and other traditional classification methods for its good performance.In this dissertation, a research is made into the principles of SVM, classification algorithms, realization, applications, and gives a SVM algorithm that can be implemented in DSP. Through simulation in CCS and debug in C6713DSK, we validate the feasibility of the method. Experiment results indicate the improvement of calculation efficiency and the feasibility in practical system. The content is as follows:(1) SVM is a theory that is based on Vapnik Chervonenkis Dimension, generalization performance, extensibility. Support vector and kernel function are important concepts in SVM, and we introduce them in chapter 2.(2) Analysis and research on SVM algorithm: There are many SVM algorithms that are based on SVM theory. From the classical classification algorithm, chunking algorithm to decomposition algorithm and the algorithm derived from above. We introduce them in chapter 3.(3) We research on the hardware structure and software development on DSP, and give a discussion for implementation in DSP(4) Speaker recognition based on SVM is another important topic. In order to improve the performance of speaker recognition, the key techniques are feature extraction and recognition algorithm. MFCC coefficient and SVM Algorithm are used in this thesis.