End-to-end Donaldson matrix estimation from fluorescence emission under programmable multi-illumination excitation using a hybrid CNN-transformer model

Donaldson matrix serves as a fundamental tool for fluorescence characterization. Currently, conventional acquisition of the Donaldson matrix via double-monochromator method is time-consuming, expensive, and inefficient, limiting its widespread application. To address this, we introduce a new measurement paradigm that shifts the complexity from exhaustive hardware scanning to intelligent computational inference. This can be achieved through a synergistic system coupling a digital micromirror device (DMD)-based programmable light source with a dedicated CNN-Transformer model. The system computationally reconstructs the complete Donaldson matrix from a minimal set of just 5–6 spectrally rich excitations in an end-to-end manner, eliminating the need for complex preprocessing. Quantitative evaluations under diverse combinations of broadband and narrowband light sources demonstrate robust performance across varying excitation conditions. The model generalizes well to samples with single and dual fluorophores, covering near-UV to visible excitation ranges, maintaining median RMSE values within 0.03 in both reflectance and fluorescence dominant regions. Additionally, the model's generalization is demonstrated by its ability to accurately estimate analogous spectral radiance factor matrices under various standard illuminants. By anchoring the model to ground-truth data from a reference double-monochromator, our approach ensures full metrological traceability, achieving an expanded uncertainty of 0.062 (k = 2) as per GUM guidelines. The proposed end-to-end measurement paradigm streamlines fluorescence characterization by directly estimating the Donaldson matrix from raw spectral inputs. Its generalization across various illuminations and samples with diverse spectral features suggests significant potential for practical fluorescence analysis under a wide range of experimental conditions.