Determining the number of pure chemical components in the mixed spectral data based on eigenvalue sequences transform

Determining the number of pure chemical components is an important step for various chemical data analysis methods like cluster analysis, principal component analysis, and spectral unmixing. In this paper, a method of eigenvalue sequences transform is proposed to improve the performance in determining the number of chemical components in spectral matrix. The proposed method converts the spectral data cube to eigenvalue sequences by applying the singular value decomposition technique firstly. Then, the method innovatively transforms the normalized eigenvalue sequences into a redefined coordinate system and detects the number of chemical components by searching the sequence of the highest point. Since the proposed method identifies the number of chemical components from the angle of geometry, all processes need not involve the use of time-consuming iterations, extensive calibration tables, or pseudostatistical hypothesis. This paper also evaluates the applications of the proposed method with simulation and real-world spectral data. The evaluation results show that the method has stronger robustness, better accuracy, and higher automaticity in estimating the number of chemical components by comparing with some calibration methods.