Research On Target Recognition And Classification Based On Data Mining Technique

As a promising and flourishing frontier of the Information Technology, Data Mining acts as the essential procedure in knowledge discovery in database (KDD). And the Support Vector Machine (SVM) is a new kind of learning machine, which is based on the Statistical Learning Theory. Its complete theory and excellent performance make a potential future for mine intelligence. The aim of this thesis is to realize 3-class underwater targets' recognition by means of Data Mining technique. The main work and originality in this thesis can be summarized as following:1. Analysis of acoustic characteristics of ship's radiated noise both in time and frequency domain, especially its deviation from the ordinary statistical signal. Studies on two modern signal processing methods-Higher-Order Statistics (HOS) and wavelet transform.2. A novel signal processing method based on alternate feature optimization is introduced and analyzed in this thesis. And a new underwater target recognition system using the optimized feature and SVM is presented here. The system utilizes the alternate feature extraction method to optimize the feature selection process.3. The optimized feature set feeds a 3-class classification module, which is based on the traditional binary SVM classifier. And the proposed linear programming SVM reduces the burden of the SVM classifier and improves its learning speed and classification accuracy. A new algorithm that combined SVM with K Nearest neighbor (KNN) is presented and it comes into being a new classifier, which can not only improve the accuracy compared to sole SVM, but also better solve the problem of selecting the parameter of kernel function for SVM.4. The numerical experiments show that the proposed data mining system based on SVM can achieve effective results in underwater targets classification.