Fused Lasso Logistic Matrix Regression Method for Multi- Source Data Fusion of Raman Spectroscopy

Abstract Raman spectrometer has a broad application prospect in the identification of dairy products quality. The diversity of measuring instruments makes the spectral data present multi-source. Multi-source data can provide more accurate information than single-source data by supporting, supplementing and correcting each other. How to fuse spectral data from different sources is a valuable research problem. In this paper, the Fused Lasso Logistic Matrix Regression (FLLMR) model is proposed for multi-source spectral data fusion, and a fast-iterative algorithm is designed by using Auxiliary Problem Principle (APP). The iterative process of the algorithm has a simple closed form. In order to verify the validity of the model and algorithm, we used 10 spectrometers to collect Raman spectral data of two kinds of dairy products respectively. Then, we used FLLMR method to perform fusion and discriminant analysis on the collected data. Experimental results show that the FLLMR method can effectively process multi-source Raman spectral data, and the model results have good interpretation. On the other hand, experiments verify the effectiveness of the proposed algorithm.