Crude oil density prediction and visualization by CLSM fluorescence spectroscopy combined with chemometric methods

This study presents a metrologically traceable, microscale measurement framework integrating Confocal Laser Scanning Microscopy (CLSM), fluorescence spectroscopy, and chemometric modeling for quantitative crude oil density mapping. Thirty-three samples from the Songliao Basin (0.764-0.9655 g/cm³) were analyzed under 488 nm excitation with emission spectra collected from 500-794 nm. Baseline correction, smoothing, and normalization improved spectral consistency, while principal component analysis with Hotelling's T² removed two outliers prior to modeling. Support Vector Regression (SVR) outperformed Partial Least Squares Regression (PLSR), achieving a calibration R²_cal of 0.968 (RMSEC = 0.008) and a prediction R²_pre of 0.955 (RMSEP = 0.012). The 737 nm wavelength exhibited the strongest correlation (r = 0.748, Pearson correlation coefficient) with aromatic and asphaltene content, linking molecular composition to density variation. Applied to CLSM spectral images of shale thin sections, the SVR model produced micrometre-scale density maps consistent with compositional heterogeneity-light fractions along fracture walls and heavy fractions within fracture interiors. A standalone desktop tool has been developed to enable spectral import, automated processing, and density visualization. This non-destructive, high-resolution approach addresses the limitations of bulk measurements, enabling reproducible, spatially resolved density determination in complex geological matrices and advancing measurement science.