Interpretable Deep Learning for Core Permeability Prediction and Upscaling

This dataset is the supporting data for the article entitled "Interpretable Deep Learning for Core Permeability Prediction and Upscaling". Sedimentary core samples used in this study were collected from Xinjie Town, Yijinhuoluo Banner, Ordos City, Inner Mongolia Autonomous Region, China (39°20′ N, 109°30′ E, elevation: 1249.70 m) .Tectonically, the study area falls within the Dongsheng Uplift-Yishan Slope tectonic belt, a vital structural unit in the northern Ordos Basin. The target formation of drilling belongs to the Permian System, with the drilled interval ranging from 0 m to 2250.00 m. Two types of specimens with distinctly different pore structures were selected for this research, including two low-porosity mudstone samples (LP) and two high-porosity coarse sandstone samples (HP). A large industrial XCT scanner (ZEISS METROTOM 1500, 225 kV G3) was employed for scanning under multi-scan mode at 220 kV and 215 µA. All scanned images were obtained at a spatial resolution of 43.43 µm. To balance exposure level and signal-to-noise ratio (SNR), the gain was set to 3.0 and the integration time was fixed at 1000 ms.This dataset contains raw XCT-scanned rock core images and post-processed data. Dominated by clastic rocks including mudstone and sandstone, this formation is critical for regional energy exploration and engineering construction. Therefore, investigating its petrophysical properties and pore structural characteristics carries important scientific research value.