Researchers have proposed an Efficient Adaptive Aggregation Network (EAAN) for two-dimensional digital rock image super-resolution reconstruction. Described in Earth Science Informatics, the newly introduced architecture tackles the persistent challenge of accurately recovering fine structural features from low-resolution geological images without imposing unsustainable computational burdens.
The investigation was motivated by recognized shortcomings in contemporary deep convolutional neural network methods commonly applied to digital rock super-resolution tasks. In recent years, convolutional neural network approaches have achieved measurable progress in reconstructing mineral composition details and pore structures from lower-resolution inputs. Nevertheless, these models encounter persistent performance hurdles when processing complex rock morphology. The inherently local receptive field of CNNs struggles to accurately reconstruct complex microstructures such as pore networks and tortuous throats.
Beyond the spatial limitations of local receptive fields, prevailing deep-learning architectures for digital rock imaging suffer from severe computational inefficiencies. Most existing architectures are built with excessive network depth, resulting in high computational costs and limiting their deployment on resource-limited devices.
To achieve this balance, the authors designed EAAN as a hybrid CNN-Transformer cooperation framework that jointly optimizes local feature enhancement and global dependency modeling. The internal network architecture integrates a dual branch residual attention block, abbreviated as DBRAB, which incorporates an embedded linear multi-scale semantic attention module, designated as LMSA. Working alongside these convolutional components, a global feature adaptive transformer, or GFAT, models wide-ranging spatial dependencies across the entire image domain. This architecture maintains an efficient lightweight design.
The research team validated the performance of the EAAN architecture through rigorous benchmarking on the DeepRock-SR 2D datasets, comparing its outputs against contemporary super-resolution methods. The experimental results demonstrated a clear advantage for EAAN, which successfully achieved simultaneous improvements in reconstructed image fidelity and reductions in overall computational demands. Crucially, the authors extended their evaluation beyond standard visual metrics, confirming that EAAN accurately preserves critical petrophysical properties across the reconstructed images. This provides reliable data support for downstream reservoir characterization.
Subject of Research: Digital rock image super-resolution reconstruction
Article Title: EAAN: Efficient adaptive aggregation network for digital rock image super-resolution
Article References: Sun, Y., Li, Y., Chen, R., Liu, Y., & Liu, C. (2026). EAAN: Efficient adaptive aggregation network for digital rock image super-resolution. Earth Science Informatics, 19(12), Article 242. https://doi.org/10.1007/s12145-026-02303-7
Image Credits: AI Generated
DOI: 10.1007/s12145-026-02303-7
Keywords: digital rock imaging, super-resolution reconstruction, deep convolutional neural networks, Efficient Adaptive Aggregation Network, dual branch residual attention block, linear multi-scale semantic attention, global feature adaptive transformer, hybrid CNN-Transformer, pore structures, mineral composition, pore networks, tortuous throats, local feature enhancement, global dependency modeling, DeepRock-SR 2D datasets, petrophysical properties, reservoir characterization, image fidelity
News Source: Violet Maxwell. (November 12, 2026). Hybrid Network Improves Digital Rock Image Resolution While Preserving Critical Petrophysical Properties. Scienmag.



