The characterization of Brazilian Pre-Salt reservoirs traditionally relies on methods such as well logging, geophysics, and petrography. While effective, these techniques can be time-consuming and potentially destructive. Hyperspectral imaging (HSI) in the short-wave infrared (SWIR) range offers a rapid, non-destructive alternative for high-resolution mineral mapping, yet its application to Pre-Salt carbonates remains underexplored. This study presents a novel workflow utilizing HSI to conduct a mineralogical and diagenetic analysis of drill core samples from the Santos Basin.

Hyperspectral data (930–2500 nm) were acquired using a HySpex SWIR-384 camera, yielding a spatial resolution of 0.22 mm/pixel. A 1D Convolutional Neural Network (1D-CNN) was employed to generate detailed mineralogical maps. The methodology successfully distinguished mineralogical associations, lithotypes, and facies. Key results include the discrimination of dolomites with varying Ca/Mg ratios—validated by XRF—and the differentiation between crude oil, bitumen, and paraffin.

Critically, the analysis delineated three distinct diagenetic phases within the core sequence: (i) calcite cementation at the base; (ii) dissolution and dolomitization in the middle section; and (iii) complete calcite dissolution at the top, followed by precipitation of Mg-rich dolomite. These findings align with established diagenetic models but were achieved orders of magnitude faster than traditional techniques. This research demonstrates that HSI, processed through machine learning, is not only effective but also highly efficient for the detailed characterization of complex carbonate reservoirs, providing a robust framework for rapid core screening and offering critical insights into diagenetic history, thereby enhancing reservoir modeling efforts.