This thesis is focused on automatic classification approach and processing software system building of the spectra data. The main points of this thesis are summarized as follows: A novel class-discriminant method – the covering algorithm. The method is composed of the following two steps: the classification problem is first converted into a set covering problem, then the classification is carried out by solving the support covering sets. The key point in the method is to solve the minimum support covering sets, and four algorithms are proposed based on the minimal/maximal distance between clusters, the maximal distance of class and the local optimization. By applying the covering algorithm to the classification of normal objects versus emission-line objects, normal galaxies versus stars and starburst galaxies versus Active Galactic Nuclei, the experimental results on simulated spectra and real spectra show that the proposed covering algorithm is of high accuracy and good robustness. The classification method for stellar spectra based on Non-negative Matrix Factorization (NMF). NMF is a novel data analysis method which is effective and widely used in the processing of high dimensional data. The original data matrix, basis matrix and feature data matrix of NMF are all non-negative, so the basis of feature space has physical meaning for the dimensionality reduction of spectra. The experimental results of more than 70,000 stellar spectra from SDSS DR3 show that the NMF based classification method is of high accuracy and good robustness. Design and implementation of the LAMOST automatic/interactive processing subsystem. The automatic processing subsystem design includes: software architecture, user interface design, main function module design and classifier design. Considering the system efficiency, all the key supporting algorithms are implemented in standard C/C++. In addition, structural modularity is taken into account for further system updating and extension.The interactive processing subsystem possesses the following functionalities: PCA analysis, spectra preprocessing, spectra recognition and redshift measurement. It is composed of two structural layers: system interface layer and algorithm module layer. The buttons and menu are used for user interface and DotNET is used for developing the interface layer with the development language C#. The codes of the algorithm modules are programmed in C/C++ and compiled into DLL. The system can be operated by transferring DLL through the user interface. The system design and implementation for the automatic processing and management software system. Based on C/S and B/S, a mixed software architecture of C/S and B/S is adopted. White/black box testing technique is used to test the software by a large numbers of designed testing cases. And the test results show that the implemented software system can well satisfy the requirements of LAMOST.
修改评论