Quantitative Characterization and Evolution of Nanopores Hosted in the Secondary Organic Matter of Shale Reservoirs Based on the Improved U‑Net Approach.

Nanopores hosted in secondary organic matter (SOM) serve as critical spaces for hydrocarbon storage and migration within marine shales. Scanning electron microscopy (SEM) could enable direct visualization of nanopore structures within shale reservoirs, but analyzing large volumes of data solely through manual methods is time-consuming, labor-intensive, and highly subjective. Using SEM image data from Wufeng-Longmaxi Formation shale samples with varying maturity in the Sichuan Basin, this study proposes an improved U-Net model integrated with a multiscale input module and a hybrid attention mechanism for high-precision automated segmentation and quantitative characterization of SOM-hosted nanopores of shale reservoirs. Compared with the standard U-Net, the improved model shows significantly higher segmentation accuracy (porosity error of 0.10 vs 0.16), especially for small pores, and improves boundary delineation while reducing misclassifications from mineral grains. This enables rapid, reproducible, and large-scale quantitative analysis of nanopore networks. In the studied shale samples, SOM primarily occurs in the interparticle pores of rigid grains, intraparticle pores of clay platelets and pyrite framboids, and the associated nanopores display predominantly bubble, irregular, or spongy shapes. Moreover, the SOM-hosted nanopores are relatively sparse in the studied shale samples with the relatively low maturity [Equivalent vitrinite reflectance (EqVR