Composite classification methods on composition identification

Composition identification is an important topic of science research. With the help of spectral analysis, it can be completed much faster. However, the effectiveness of spectral analysis highly depends on reliability of reference spectrums and similarity measurement formulas. To overcome main obstacles of spectral analysis, the paper presents new concept of composite classification and three fundamental methods, Direct Similarity, Feature Series and Weighted Feature Series. Firstly these methods involve discretization and reduction in help lifting precision and reducing computational complexity. Then they compute similarities by their own criterion separately and finally make judgments to give out results. The experimental results prove the effectiveness and efficiency of these methods on composition identification of real world dataset.