Fluorescence excitation-emission matrix (EEM) spectroscopy is a crucial analytical tool for characterizing dissolved organic matter in aquatic systems. The factorization of mixed spectral components within EEMs has long been the main subject of data interpretation, prompting widespread adoption of trilinear decomposition such as parallel factor analysis (PARAFAC). However, the requirements of multi-sample dataset and manual judgment pose limitations to PARAFAC analysis, particularly hindering the real-time and in-situ applications. This study introduces a rapid decomposition approach capable of automatically decomposing single EEM input into fluorescent components. The proposed approach, termed empirical initialization non-negative matrix factorization (EI-NMF), comprises three core steps: (1) chemical rank estimation via singular value decomposition (SVD), (2) empirical initialization based on statistical analysis, and (3) non-negative matrix factorization with multiplicative updates. Simulated data and natural water samples were used to verify the feasibility of proposed approach. Validation on simulated data yielded satisfactory results: EI-NMF achieved accurate chemical rank determination and component spectral recovery (Tucker congruence coefficients >0.9) relative to the true component spectra. Decomposition results of unseen natural samples further confirmed that EI-NMF can effectively processes single EEM inputs, yielding decomposition outcomes with excellent accuracy and chemical interpretability. This computationally efficient framework enables real-time decomposition of individual EEMs (processing time <0.1 s), offering significant potential for in situ monitoring of aquatic fluorescent components.