Font Size: a A A

Curvelet And Fast Sparse LSSVM Based Target Recognition

Posted on:2011-05-07Degree:MasterType:Thesis
Country:ChinaCandidate:Y F ZhangFull Text:PDF
GTID:2178360305964097Subject:Circuits and Systems
Abstract/Summary:PDF Full Text Request
At present, kernel machine has been applied in many fields and plays a more and more important role. In the past few years, the scale of the data we got becomes larger and larger, which leads to higher computational complexity. In order to solve this problem, many people researched on sparse kernel machine. Synthetic Aperture Radar (SAR) is a new important subject developed in recent years, as the key composition in SAR image interpretation and analysis, SAR target recognition is the research focus in the fields of SAR image processing and pattern recognition. In order to enhance the performance of SAR target recognition, this paper focuses on the research on kernel machine learning and Multi-scale Geometric Analysis in SAR target recognition, and proposes some new methods. The main contributions can be summarized as follows:(1) In order to obtain the sparseness of Least Square Support Vector Machine, an Improved Fast Sparse Least Square Support Vector Machine (IFSLSSVM) is proposed and applied to SAR target recognition. The proposed algorithm combines incremental and decremental learning, selects the important samples from the training set as support vectors, and then iteratively constructs the decision function. Benchmarking UCI datasets are firstly used for validating the performance of the proposed algorithm, followed by SAR target recognition. Experimental results on MSTAR SAR dataset show that IFSLSSVM is an effective SAR target recognition approach, which not only reduces the number of support vectors but also enhances the recognition accuracy.(2) Combining multi-scale geometric analysis and machine learning, a Two-Dimension Curvelet Kernel Least Square Support Vector Machine is proposed. The construction of two-dimension Curvelet kernel, which provides more option for support vector kernel, is based on the conditions of support vector kernel and the advantages of Curvelet, like anisotropy scaling relation. The effectiveness of Curvelet kernel is verified by the experiments of classification and function approximation.(3) A new SAR target recognition method is proposed, which combines Curvelet analysis and KFD as the feature extraction technology and employs LSSVM as the classifier. The proposed method firstly uses Curvelet to extract the low-frequency information of SAR image, reduces dimensions by KFD, and then implements recognition by LSSVM. The experimental results show that the proposed method can receive better recognition accuracy, which means it is an effective SAR target recognition method.
Keywords/Search Tags:Synthetic Aperture Radar, Sparse Least Square Support Vector Machine, multi-scale geometric analysis, Curvelet, support vector kernel, feature extraction KFD
PDF Full Text Request
Related items